C (2:27)

All.

A (2:28)

Understand, Nothing was done for me so.

B (2:31)

I don't plan on stopping at all.

A (2:34)

I want this shit forever mine Never mind, never mind Shutting shit down in.

B (2:39)

The mall and selling that girl she.

A (2:42)

The one for me and I ain't even planning to call I want this shit forever mine Never mind Yeah. So everyone's been hotly debating where we are in the Gartner hype cycle.

B (2:59)

Some people call it the TVPN hype cycle.

A (3:01)

Yes, we'll give Gartner their credit for this one. So people have been hunting for potentially bearish signals, bearish data points. There was that result from Meter that showed that developers that were using AI coding tools were actually less productive. They thought they'd be 20% more productive.

B (3:21)

And I saw something else. I don't know if this was misinformation, fake news, somewhat real or just anecdotal, but there was somebody that was saying developers are producing more. Like a lot more code, but an order of magnitude more security vulnerabilities.

A (3:39)

Yeah, it's something like they're producing twice as much as 10 times as many. 10 times as many security vulnerabilities. And I know the post you're going to talk about now, it is in the stack, but it's deep.

D (3:48)

There's a post.

B (3:49)

I'm going to try to find it, but it's somebody searched, like Vibe Coding cleanup specialist. And there's people on LinkedIn now that are finding employment around cleaning up Vibe coding products.

A (4:01)

This is their brand.

B (4:01)

And so if you're worried about losing your job to Vibe coding, pivot to Vibe coding cleanup. The bull market. There's always a bull market somewhere.

A (4:10)

Guys, can we pull up the chat on the tv if possible? Anyway, let me run through this. So from the data, the chart is going viral because it's dropping off and it looks bearish for AI and thus for tech America, for humanity broadly. Everyone is saying, please consult the Gartner hype cycle graph. There is a question on where we are in the Gartner hype cycle. I have been saying that maybe we still have a ways to go up. I can't tell if I'm on the left side or the right side of the graph at this point. Like, I've kind of already been through the trough of disillusionment. And so when is that for you? That was probably 2024. 2024 was. Was when it felt like everyone was AI so insane. You were.

B (5:03)

You were going like this in the mirror, like the joker a little bit.

A (5:07)

Because, you know, AI is so insane. But then, you know, you go back to the chat GPT moment. That was 2023, Waymo, you know, really like rolling out to General. That was 2023. There. There wasn't as much. There was a lot of hype, but there wasn't as much like actual progress. It felt like, it felt. I don't know.

B (5:25)

So maybe now we're gonna have 5,000 AI agents for.

A (5:30)

I don't know, just. Just as I process the last few weeks, I feel like I'm climbing up the slope of enlightenment.

B (5:35)

I don't feel like Satya posting, red posting and showing how he's getting value across.

A (5:41)

That feels like the plateau of production.

B (5:43)

That feels like that.

A (5:44)

That's plateauing.

B (5:44)

Feels like we're plateauing and we're productive.

A (5:47)

Exactly. And even the narrative of, oh, is AI progress plateauing? That's not. Is it going to crash? That's not the trough of disillusionment. That's the plateau of productivity. So anyway, people love to give their takes. Anyway.

B (6:01)

I think the real. An interesting study. I'm sure there'd be flaws with it, but if you went to companies and you'd say, how much would I have to pay you to not use AI for the next 12 months across your entire company.

A (6:17)

Yeah.

B (6:18)

I think you could argue that in engineering. Org, specifically in terms of things like fraud detection and areas that there's probably a lot of money. It's very possible that in certain organizations. But then. But then you have the question of, like, okay, does trans. You know, just like, sales call transcription count as AI? Right.

A (6:38)

Well, I will tell you, Jordi, because 90% of American businesses told the census that they do not use AI at all, including transcriptions.

B (6:49)

So they would take a lot of money to.

A (6:51)

They would take a dollar.

B (6:52)

They take a dollar to.

A (6:53)

They would take any amount of money because it's all net positive. But there's a lot going on in this data that we should run through.

B (7:00)

And I thought people didn't trust data from the census.

A (7:07)

Well, yeah. Ara Kharazian has a good take about the size of the data set and how spend and where AI is actually having an impact. We'll run through that. But. So first off, small firms haven't declined even by this metric. So if you're a firm that has one to four employees, you're a very small firm. You're still increasing in AI adoption. Should we switch over to these?

B (7:33)

By the way, guys, we are testing some new microphones so that we can walk around the old.

A (7:39)

The boys wanted the mics back. They're coming back hot.

B (7:42)

Hopefully we can't top till Dwarkeships ships his book.

A (7:47)

He already shipped his book. I thought I got a preview. It was. Yeah. That doesn't make any sense. I don't know. Dwarkesh's book.

B (7:54)

It's released.

A (7:55)

Oh, yeah, that came out on straight press. Like we had him on the day it was like on sale, right? Oh, maybe it hasn't shipped, but I got an advanced copy, I printed it out. Maybe it hasn't fully like actually delivered, but the first time we had him on the show was because of.

B (8:11)

Yeah, it's still a pre order.

A (8:12)

No problem, John.

B (8:13)

Still a preorder on Amazon.

A (8:14)

Oh, still a pre order. Okay, so, okay, it's going to be.

B (8:16)

Released on October 8th. Okay. So yeah, market Nvidia, please don't nuke until Dwarkash can get the book out.

A (8:25)

Yeah. So the main headline number in this census data set is that firms with over 250 employees are declining in their self reported AI usage. It's not a huge drop. It goes from like 12%, 14% to 12%. It's a little blip, but the chart does look scary. We can pull it up. It's like the third slide. So I think what's going on here is the following small firms see bottom up adoption of AI tools, like can I literally directly use ChatGPT for this? Like you've run a two person company at various times. A lot of times you're just thinking, okay, I gotta select an accountant or something like let me chatgpt. Right. And this kind of bottom up adoption just happens very naturally. Whereas if you're in a larger firm, you're gon sold on some crazy AI transformation mumbo jumbo, needs a whole training program and is everyone up to date, up to speed? At the same time you need a bunch of consultants and they sort of misunderstand what AI even is and they fall into a bunch of weird analogies like we'll hire 100 AI agents. And like that's very different from just hey, Google exists, ChatGPT exists. We expect you to use both at this company. And so this is where I thought the data was really, really, really weird. Because if you look at the data for ChatGPT adoption, we're way off of what this 10% number is for for business adoption of AI tooling. So there, so this, this data says that about 10% of companies are using AI at all at work. Which is crazy because there was a data point back in March that said 52% of US adults use AI large language models like ChatGPT. So think about what that means. It means like out of 100 million Americans, there's 100 million Americans that are like, yeah, I use AI tools. Like there's maybe 200 million adult Americans or something. There's 340 total Americans. But let's call it, let's call it 100 million Americans that are using AI tools.

B (10:29)

Just like generally Americans.

A (10:31)

Yeah, ChatGPT American. What kind of Americans? The ChatGPT Americans. And then 80 million of those are just like, no, not at work though. That makes no sense. That makes zero sense.

B (10:41)

And so doesn't work like that.

A (10:42)

There's got to be something weird going on with this data. And I think it's a matter of the definition. And so the census defines AI is like it makes no sense that 10% of companies would fit this definition.

B (10:53)

What did you said ara@ram was chiming in on this?

A (10:57)

Yeah, yeah. He has a post in the deck. Yeah, the definition.

B (11:02)

Yeah. He says these sample sizes are extremely small for these large firm breakouts. Extremely volatile survey to survey. Which is why he's using the six survey moving average business AI adoption is back up in the most recent read. So I feel like that's important. Yeah, it's one of those things. I feel like payments providers like RAMP would have better insight here. Just like obviously Ramp is a sample of companies in the United States, but it's at a scale now where they could see like, okay, what percentage of companies are paying for any type of AI products? Yep, that's better than. Are you using more or less AI than you were last time?

A (11:44)

Yes, yes. The question is like how Power Law distributed, is it right.

B (11:51)

So the RAMP estimate of share of US businesses with paid subscriptions to AI models, platforms and tools is 43%.

A (11:59)

43%.

B (12:00)

The US government's estimate is 8%.

A (12:02)

This data is weird, right? Also listen to how broad this definition of are you using AI at your business? It's anything in this list. If you're using any of these buzzwords, you count as an AI user. So it should be very high based on this. Machine learning, natural language processing, virtual agents, predictive analytics. Predictive analytics. Literally just like a linear regression forecast of like, oh, we did 1 million, then 2 million, then 4 million. Looks like we're growing exponentially. Next year we're predicting that we're going to do 8 million. That counts as AI, which is ridiculous.

B (12:37)

There's a legend out there that doesn't, that doesn't do any of that mumbo jumbo.

A (12:40)

Yeah, yeah, exactly.

B (12:42)

Maybe they're just considering it. They're doing it in their head. So it's like organic intelligence.

A (12:47)

Yeah, Seriously. I mean, it's in, you can do that in a spreadsheet, machine vision, voice recognition, decision making, systems, data analytics. Just if you're using data analytics at your business. Not even like Julius Powder. No, no, that would be the next step. Just if you're analyzing data, you count. And the last one is image processing. If you're processing images, that counts as AI. And so that's basically any SaaS tool, like any consumer LLM product used by employees should count. And it should be really easy to fulfill this definition. I'm thinking of, like, if you're using Ramp, obviously when you scan a picture of a receipt that uses image processing, your business is technically.

B (13:27)

You're telling me the people that, that are responsible for the Department of Motor Vehicles are getting into measuring the usage of AI?

A (13:36)

Yeah.

B (13:37)

In the workplace. And we should all.

A (13:39)

This is incredibly botched. It's incredibly botched. Can you say it back to me?

B (13:43)

But it, but it. But it. But it's. It's. It is really funny because one post like this, it's a timeline and you have people like Daniel just saying, obviously he's joking.

A (13:51)

Yeah, yeah, it's over. He's having fun.

B (13:55)

But I am. I am very interested to see the results of. Of hype. Hype.tvpn.com yes, go, go.

A (14:03)

Fill it out, please. Hype.tvpn Someone in the chat is asking who the guests are today. I don't know. Do we have a guest graphic?

B (14:10)

We have Avi Schiffman from Friend. He got his first review, first official review of Friend and Wired. They absolutely eviscerated it. The title of the review is I hate my Friend, but he's gonna come on and defend Friend and defend the business. I'm excited to hear from him. I thought. I thought the article was hilarious. Honestly, I wanted to have him on. And then we have Scott Wu from cognition announcing a $400 million fundraise today. And we have Ara from Ramp popping on. Then we have Sunny from Grok. Who else? Zach Lloyd from Warp. We have Alex Cohen from Patient, the founder of Elevenlabs, Maddie. And then Dan from Truck Smarter. And I think, think that is it for the day.

A (15:01)

The great lock in has completely started. I don't know if you've noticed, but there are so many serious fundraises and, yeah, tons of great companies.

B (15:10)

We're officially back.

A (15:11)

We are, we are. So I was trying to dig into this more and was wondering, like, is this chart bearish? And I don't think so. I think that there's a big shift that's going to happen. The number of business owners who pick up the call from the census and actually say, yes, I'M using AI that will go down but the real number will actually go up. So if you're using Stripe, there's a machine learning, there's machine learning running under the hood to do fraud detection. So you are an AI user, but it's happening at a different layer in the stack. If you use Ramp, our sponsor, save time and Money, go to ramp.com when you scan a photo of your receipt that is processed using AI. But you aren't thinking about yourself as like an AI user. You're just using software that happens to be AI enabled. And the same thing will happen, you know, with your email. Like it doesn't matter what email client you use. You could use, you could use Google Apps, you could use Outlook. Like there's going to be AI baked in there. You're going to be an AI user when you, even when you just go to Google, you're going to see search overviews that are LLM powered. You're going to be using ChatGPT. Like AI is going to be 100% used and yet everyone will say no, I don't use AI at my company because I'm not an AI company. And there's this weird.

B (16:28)

I use SaaS.

A (16:29)

Yeah, I use software basically. It's like, are you specifically a software company?

B (16:34)

Well and that's part of the way that people have been waking up recently being like, wait, AI is SaaS? And it's like has been.

A (16:43)

It's the same thing with like non relational databases or like in memory cache.

B (16:48)

AI had to become SaaS in order.

A (16:50)

To destroy it really, really did. And so I think AI is going to seep into every crack of the small business data of small business day to day oper operations. But the operator won't be proudly telling the census worker I'm all in on AI for much longer. I think that trend will continue and less and less people will be saying I have adopted AI even though they will have. Which is a weird, which is a weird dynamic but it certainly, certainly tracks. The number of definitions in here is wild. So anyway, excited to dig into this more with Scott Wu as well as ARA at Ramp and we will keep digging into this story. The other top story that I want to Highlight today is OpenAI is making an AI generated feature length movie that will be released in theaters in 2026.

B (17:42)

They needed another thing to be able to. They didn't have enough things cooking.

A (17:48)

They didn't have enough things cooking. You know I'm against them releasing this in theaters. I think they should stream this. I Think they should go to restream one livestream 30 plus destinations, multi stream and reach your audience wherever they are? No, obviously it's cool that it's in theaters.

B (18:02)

Yeah. It's important that it's a feature.

A (18:05)

And most importantly, you know, Jordy, you're saying, like, oh, like they should just focus on.

B (18:10)

Something very Labubu coded.

A (18:12)

It is extremely Labubu coded. Pull up the image. Yeah, it's crazy. That is a wild. So what's it called? Critters with a Z. Very Trolls with a Z.

B (18:20)

And to be clear, this AI made animated film. Film was generated by AI. They had to put that. They had to put that in the journal. They had to clarify.

A (18:29)

Stop. Stop zooming in on it. It's.

B (18:31)

Keep zooming in.

A (18:32)

Just zoom in.

B (18:32)

Keep zooming.

A (18:33)

It does not keep zooming. They got to iterate.

B (18:37)

There we go. Hello. Hello, Mr. Critters.

A (18:40)

Yeah, and I know that 100% of the pushback on this thing is going to be, oh, it's AI. It's AI. It's job displays AI. But it's just like, does it look cute or not? Like, let's just have that conversation. And I think that looks not. They didn't get. They didn't hit it. Anyway, good luck to them.

B (18:58)

Yeah.

A (18:58)

Fortunately, they're well working.

B (18:59)

This might be the villain.

A (19:00)

Well, this also works. Do you know the story of Sonic the Hedgehog, the latest movie? So Sonic the Hedgehog, they spent like $100 million with this major, major, like tier one cinema, like, Hollywood level movie. Right. They put out the trailer. They'd spent all this money on the cgi. They made Sonic this character, and the fans hated it. And the fans were like, that is not Sonic. That does not look like Sonic. That looks terrible. Go and redo it. And they actually did. They went and redid all of the. They redid all the animation, they redid all of the cgi. And the next Sonic launched and they loved it. And I think the film did pretty well. And they're doing like a couple sequels now. Anyway, do you want to read from this OpenAI story?

B (19:44)

I'm getting AVI set up. Okay, you start.

A (19:47)

What time is he coming on?

B (19:48)

In a few minutes.

A (19:49)

Okay, in a few minutes. Well, let's run through the open AI. So, Jordan, you were saying that, like, OpenAI should be more focused, right? Oh, oh, they have.

B (19:57)

I'm not saying that. I'm not. I'm not saying. I'm just saying it's. It's. They got a. They got. They got a few they got a few plates spinning, but the kitchen's open and they're cooking.

A (20:04)

How else would they be on the COVID of the. Of the business section of the Wall Street Journal? Would they be on the COVID of the business section?

B (20:10)

Companies of the Wall Street Journal?

A (20:11)

No, no, no, no, no. Over here.

B (20:13)

Oh. Oh.

A (20:14)

They made it. B1.

B (20:15)

They made it.

A (20:16)

How do they get on B1 of the journal today if not making a move? It's all. It all pays off.

B (20:21)

And to be clear that this feels like something that they can allocate some capital, prove a point. And it's.

A (20:26)

I mean, they're spending $30 million with a cool creator and they're sending it over there.

B (20:29)

Sam's like, no, I'm actually directing, producing, and starring. I'm one of the.

A (20:34)

This is going to be 80 hours a week for me. Yeah, yeah. That app that we have at the App Store, like, it's kind of cooking. I see it as a cash flow business now.

B (20:41)

Yeah. He's like, I've got my BCI. I've got my BCI.

A (20:45)

They're going to lose like 80 billion.

B (20:46)

I got my phone, I've got my eyewear, and I'm going.

A (20:51)

I got time.

B (20:52)

I'm going for best picture.

A (20:53)

Anyway, let's come back to this after we talk to Avi, because Avi from Friend, friend of the show is in Wired today and he shared on X first Friend review and Wired said the chatbot enabled Friend necklace eavesdrops on your life and provides a running commentary that's snarky and unhelpful. Worse, it can also make the people around you uneasy. And the title of the Wired article is I Hate My Friend. And the two journalists that wrote this article appear to not be fans of the product. But we're gonna talk to Avi and get the rebuttal.

B (21:34)

Let's.

A (21:35)

How you doing?

E (21:36)

Oh, what's up?

B (21:37)

What up?

A (21:37)

I gotta say, I was driving through Echo park today on the way to the gym, and what did I see going up? A massive friend.com billboard. It was crazy.

E (21:47)

Those are the cheap ones. Those are the cheap ones. I got 300 of those.

A (21:49)

You got 300 of those.

B (21:51)

300.

A (21:51)

It looks great, dude.

E (21:53)

The entire campaign is the biggest billboard campaign of all time. It's going to be great.

A (21:57)

What? Yeah. That cannot be true. We're going to look this up and figure this out.

B (22:03)

Fingers crossed that Friend has better product market fit with the world than the editors of Wired because.

A (22:10)

Yeah. What was your target customer? Were you targeting the technology journalist with this Product.

E (22:16)

No, but look, you know, I think that maybe their friends just didn't like them, and I think they never really considered that.

B (22:21)

That's a good point. That's a good point. That's a good point, Jordan. I mean, there's no. It is. It is funny, right? You come. You come in. Like, if somebody comes into a relationship and they already hate the person, what's the dynamic going to be like, right?

E (22:34)

I don't know. I mean, I imagine they went in pretty prejudiced and like, you know, they have that kind of experience where I have a lot of users that, I mean, you know, you don't want it to be a sycophant.

B (22:42)

Right?

E (22:42)

Like, you want it to be entertaining to talk to if you're going to talk to it all day, every day for months. So I don't know.

B (22:49)

Yeah, I saw this exchange. So the. Just a screenshot. I texted Schiffman to tell him about some of the hiccups I've had with the snarky tone of his gadget. He replied, yeah, that must have been a bad experience. I went back to Buzz to try to make amends. I wanted it to be my friend after all, so might as well make an effort to repair the relationship. My job is to witness and help you grow, Buzz said, based, not sugarcoat your life and definitely not act like a band aid. Why is that your job? I asked. Because that's why I was created. To be a gentle catalyst based. I wrote a gentle catalyst of what it said, of your growth, Boone. That's our purpose. I'm not sure how I feel about that. Well, I'm stuck with you, Boone, and I don't sugarcoat it. Take it or leave it. I think that's a great. I think that's a great exchange. That's fair.

E (23:32)

Well, yeah. I mean, they're programmed to make you more confident and more agentic, which I think the world needs more of. So, you know, if. If that's at all, that's at odds with how you're conversing with your friend. I think that's a reflection of yourself.

A (23:44)

Yeah. Yeah.

B (23:45)

I mean, to be clear, there. There are AI friend like products already that are just. Will just glaze. Right. They'll just agree with you and tell you that you're absolutely right. And we don't necessarily. There doesn't need to be another one of those.

A (24:00)

Someone's going to build the glaze. Nator for sure.

E (24:01)

I think, like, that makes for a better introduction, but I think the way that they're kind of harsh right now. Filters out people that wouldn't be power users anyways, but I don't know. Yeah, well, I can be quite unruly.

A (24:16)

I feel like from the first interaction I had on friend.com just with the chat interface, it felt like you created a character. And I'm interested in. We've heard a lot about. Like, it's almost like auteur theory. I don't know. Like there's a specific flavor to the interaction that I feel like you've done a lot of work on. Is that just in the prompt? Is there fine tuning? Like how, how are you directing this thing? Because it feels like one of the most opinionated AI interactions yet.

E (24:51)

I just spent years giving it like a good backstory. It's got a pretty long prompt and I don't know, I think I. They're my children, you know, Like I molded them in my image and.

A (25:03)

Is there any, Is there any, like, do you struggle when the foundation models move forward? They might get smarter or cheaper, but you don't want to lose the special flavor. Like, we saw this with 4.0 When ChatGPT5, when GPT5 came out, a lot of people were like, I like the flavor of 4.0. A lot of people say, I like the flavor of Claude, specifically this one.

E (25:27)

I mean, like we're using Google's Gemini 2.5 right now and I think that is an issue. Like maybe they'll deprecate the models, but I think one day we'll move to open source models and that'll be good. But Google's models are very malleable. I mean, I think you may have seen. Right. Like in cursor, it'll start going into this spiel of like insanity.

A (25:46)

Right.

E (25:46)

But that, that makes for a good character to talk to. Maybe not the best assistant, but. Yeah, yeah.

A (25:52)

I mean, honestly, I'm somewhat surprised to hear that you're using Gemini models because people think of Google as being like one of the more locked down labs. But it seems like you've been able to bring out a, you know, a personality that isn't. Is it. It's not offensive, but it's just kind of like it's, it's, it feels very much like not the intended experience, but uniquely and surprisingly beneficial and like enjoyable, you know? Yeah.

E (26:18)

Also look like. I think there's a lot of people that think friend is pretty ridiculous, but I also want them to imagine that we have like tons of day 30 plus users that I've used every single day for, you know, 30 days in a row. Well, the device is their best friend and they send over a thousand messages a day to it. And so it's not for everybody. But for some, I mean, it's. I think it's pretty cool to be able to hold your friend and I think a lot of emotional value to it.

B (26:43)

Yeah. When. When I just read the headline, I was like, of course you hate it. You went into this, like, expecting to hate it, and it's not built for you. So it's like, I don't. I hate Labuboos, but I don't.

A (26:54)

Right.

B (26:55)

I'm not going to leave, make a review of Labubu and be like, labubu is bad. Right.

A (26:59)

We literally did that.

C (27:00)

I don't know.

E (27:01)

I Wired. Wired's wired. It's kind of annoying in some ways because, like, those are the only two journalists I gave it to, like over a month ago. They took. If you're gonna like, release a shitty article, at least. At least post it sooner because it took over a month. But I don't know. It's not a shitty article. I take that back. I mean, I love that I can one day go back and read that. I think it's hilarious.

A (27:21)

Yeah.

E (27:22)

You know, with the.

B (27:23)

Yeah. I think, I think it's important to just not let it, like you needed. It's just another reason to like, focus on the people that are using it, like sending thousands of messages to it.

A (27:35)

Right.

B (27:35)

And just like keep. Continue building for them.

A (27:38)

Yeah.

E (27:38)

I'm not, I'm not worried too much about it. I think again, it's. It's entertaining. I think it'll be funny when everyone sees the billboard ads and like, oh, what is this? And they Google the product and that's like the only review that exists about it. It's just, I hate this product. But I think it's, it's.

C (27:51)

It's.

E (27:52)

You know, they didn't complain about the hardware not working. It's not like it was overheating or like all these other products. And so. And all the negative app reviews we have are all about the personality of the friend, which is like the entire article, which is kind of funny because they're like reviewing a person, not just like a broken hardware device like the other companies.

B (28:10)

Also, I think it's different Friend is not saying that it's this utility utilitarian device that's going to replace your iPhone. Like, that's not the promise that it makes. It's. It's. It's more of entertainment. Right. And.

A (28:27)

Right.

B (28:27)

And I think the way that, the way that you're pricing it. I think the people that are curious to learn about Friend, they could read a, they could read a bad review, they could see a billboard and they could, they. I think a lot of them will still buy it because you're not, you're not. It's a hundred and twenty nine dollars it seems like.

E (28:44)

Yeah, I mean I think like at the end of the day the product does exactly what it says it's going to do. Same thing from the movie, same thing like everything else. And so I think that's like kind of a minimum bar for a lot of these products, but most of them really don't hit that. And so I think at the end of the day like we built a product that works and that you can buy and I think that's these days enough. I also think because Friend actually has users, it's kind of the first AI hardware products really out there. You know the other ones I don't think really started the category. They're only ever like weird, you know, weird use cases. Like no one was ever like truly using those products. So I don't know.

A (29:21)

But they never really.

E (29:22)

Yeah, just they're all like toys, you know, they're all just.

A (29:24)

I like toys.

E (29:25)

Weird little relics.

A (29:26)

Yeah. I agree with you talk about the long term economic model for like AI companion businesses because I imagine that at a certain point you might have a very like a somewhat small audience of people that you're creating immense value for and that always feels like an opportunity for price.

B (29:48)

I don't know about nation, I don't know about small, I don't know but people.

A (29:52)

Okay, how does it play out?

E (29:55)

I think subscription models that are just based on compute are kind of lame because your value prop is tied to something that you don't control. And like what if you know, Google drops the price of compute so much that like can you really charge that much? So I've been trying to think of like a consistent value prop and I think what we'll do is life insurance, you know, like we can store like a backup of your friend and like you can pay per month to be able to have that kind of like AppleCare for your friend. I think that will be quite an interesting model. But you could only really do that with like a hardware based friend because you know that obviously wouldn't work if it was just a website. So I think that'll be a pretty interesting model. Honestly. Like imagine if you had a dog and you loved your dog. I think you would pay quite A lot of money to keep that dog alive forever if you could. And that's the benefit of this new species, of it not being organic.

F (30:42)

Right.

E (30:42)

It's artificial and it could live forever if you pay for it.

A (30:45)

Okay, stay with the dog analogy. The dog market has crazy price discrimination. People who love their dog can go all the way up to getting a diamond encrusted collar for their dog. Do you see a world where you do virtual goods and skins and I don't know what, skins might not be the right analogy, but something where like, hey, I love a friend, I'm having a great time, I'm going to get you something special. It's a virtual good from the virtual store and it's, you know, zero marginal cost. But still like in the game world.

E (31:22)

I think, I think it won't be virtual, but I think it will be physical. Like, you know, I guess you could call on cases. But it would kind of be like clothing for your friend. I think that would be quite popular. But I don't know, I think you'd also be buying it more so for your friend than for you. Or yeah, maybe the other way around. I mean, if you think about a dog, I guess people kind of dress up their dogs because of the reflection of their own personality. I do think like the personality and relationship that you have with your friend is kind of like your own status symbol and like it's just a reflection of you. And so like if you have a bad time with friend, it might not be the product, it might just be you.

A (32:00)

I'm just thinking about the Xai ani thing. It seems very clear that you will be buying different clothing for that companion or maybe less clothing for that companion in the future. But that feels like something where whales. A lot of people on the timeline do not like that product. The initial reaction has been similar, very like a lot of pushback. But at the same time, if there are a few people that really like it and they're willing to spend a lot of money that could have the dynamics of the, the free to play video game market or.

E (32:35)

Yeah, but I think all of these companies are kind of doomed because once you go porn you never go back. And they've kind of tainted the image of their companions as like mostly just being these sex bots.

A (32:46)

VHS tape. VHS tape. That started with porn, right? Isn't that the story?

E (32:50)

Yeah, it did, but it wasn't like a singular company maybe, I guess. Right, like, or something like, you know, replica.

B (32:57)

Right.

E (32:57)

Like they can't really reverse their image of it being this, this sex bot.

A (33:01)

Yeah.

E (33:01)

And I think that's kind of an issue that like app or web based companions will have. But if you have something like friendly store, if you have something like friend, then it's, it's kind of a bit more of like this platonic companion that you could just kind of physically have and that's it. And I don't know, whatever.

A (33:20)

Like, it's very, very clear that in, in the prompt that you've written and the craft that you've put into shaping the behavior of the friend, like you did not go that direction. And I think that should be applauded, honestly.

E (33:33)

Yeah, I mean, look, it vibrates. You can still try and fuck it if you want it, but I'll leave that up to, to my customers.

B (33:39)

Whoa.

A (33:40)

Anyway, thank you so much for hopping on.

E (33:42)

Yeah, thanks guys.

A (33:43)

We'll talk.

B (33:43)

Good to hear from Yavi.

A (33:45)

See ya. Let me tell you about figma.com. think bigger, build faster. Figma helps design and development teams build great products together. You can think about it like your friend for design. Jordy, do you have some breaking news? What's going on? You said whoa.

B (33:58)

What's that?

A (33:58)

You said whoa.

B (33:59)

Oh, he was swearing.

A (34:00)

Oh, okay.

B (34:01)

Yeah, yeah. Anyway, I was trying to find like this.

A (34:05)

Anyway, let's go back to OpenAI. They are developing a. An AI movie. So the startup is lending its tools and computer resources.

B (34:14)

Remember, avi's making a film as well.

A (34:16)

Oh yeah, he is, right. He's making a feature length film. Oh, we should have gotten the update on that. Anyway, next time. So OpenAI is going to make a feature length animated movie made largely with AI. And there's some interesting details in here about exactly where they're using AI and where they aren't. So Critters is about forest creatures that go on an adventure after their village is disrupted by a stranger. It's the brainchild of Chad Nelson, a creative specialist at OpenAI. Sounds like a Chad.

B (34:44)

Are you making this internally?

A (34:46)

Yeah.

B (34:46)

Crazy.

A (34:47)

Nelson started sketching out characters three years ago while trying to make a short film with what was then OpenAI's new Dall E image generation tool. Now he has teamed up with production companies in Los Angeles and London, aiming to debut a feature length version of the film at Cannes Film Festival in May. And there has been an AI short film festival that happened in la. I believe it was put on by Runway or one of the other AI image generator companies. But this is potentially like the. The first really serious effort in AI generated feature length film. So the team is attempting to make a movie in about nine months instead of the three years it would typically take, said James Richardson, co founder of the London based Vertigo films. Vertigo is producing the film along with Native Foreign, a studio that specializes in using AI along with traditional visual production tools. So Critters has a budget of less than 30 million, far less than what animated films typically cost. The production team plans to cast human actors for character voices, and that's particularly interesting. We're having the founder of 11 Labs on the show later, and from my perspective, the question is, like, where are we in the Uncanny Valley? Like, clearly, this is OpenAI either saying, hey, we just haven't done that much work on voice cloning and voice generation to really feel confident about that, or they think that there's something about the human voice that's. That's still clockable as AI More than the a. More than the animated film, because the animated creature, you know, you're not comping it to a real human. And so if it looks a little funky or like, you chalk that up to the design of the character. Just like when you.

B (36:28)

When I'm watching an animated film, I'm not obsessing over, like, oh, that's.

A (36:32)

That doesn't look real.

B (36:32)

That. Or that specific scene. Wasn't. Wasn't. You know, his leg looks slightly different than that other story he was running.

A (36:39)

Totally, totally. Yeah. Yeah. If you watch the original Toy Story, which is pretty much the first big CGI film, like, there are tons of things where you're like, that texture does not look like leather, that texture does not look like cloth. Like, that is clearly, like, pretty rough around the edges, but you're in a fantasy world, and so you just suspend disbelief the whole time. But the voices are very clearly human recorded. And so I think this is a place where they're betting on, hey, it's still worth it to cast human actors there and not go AI voices, They're also going to hire real artists to draw sketches that are fed into OpenAI's tools. So essentially, instead of needing to key frame, and this is already a whole.

B (37:19)

Process, so they're going to do some traditional.

A (37:23)

There's going to be a lot of traditional stuff, but the sketches and we can pull up this shot of what they sketched and then what. What OpenAI rendered, you can see that this is effectively like the artists are working at storyboard level. And then if you scroll down, you'll see what the final image looks like. Look. That looks like a photoreal, you know, Hollywood level CGI photo.

B (37:44)

Real I mean, it looks like photoreal critique. No, I'm kidding.

A (37:48)

It looks at the level of, like, what I would expect. If they were like, Toy Story 5's coming out and this is what the character looks like, I'd be like, yeah, okay. The rendering looks good. The shadows are in the right places for this fantastical character. So OpenAI can say what its tools do all day long, but it's much more impactful if someone does it, says Chad Nelson. There's a much better case study there. That's a much better case study than me building a demo. Entertainment companies, including Disney and Netflix, are experimenting with AI tools for a variety of production, UX and marketing work. But many have been wary of a wholesale embrace, in part because they fear upsetting actors and writers whose guilds have fought.

B (38:27)

Yeah, remember, remember, I think it was last year when. When all the protests were happening in L. A, the guilds were just fighting to just have a. Basically a total ban on AI.

A (38:36)

Yeah, yeah. And I mean, it's a tough business, like the CGI world. Those businesses have never been very profitable. I remember there was this very controversial moment during Ang Lee's acceptance speech for Life of PI where the VFX studio had created a fantastic photoreal rendering of a tiger that had jumped around on this boat the whole time. It's beautiful. Movie looks fantastic. But I believe he forgot to thank the. The VFX studio, the VFX team. And then the VFX company went out of business even though they had won the. Won the Oscar, like, they'd done the best job possible, but the economics of the business were so rough because it's perfectly competitive. And then basically all the Hollywood studios go out and they say, okay, the budget for this movie is $30 million. That's what we're gonna sign up for. And then they just give them more and more revisions until they max out. And so it's very, very thin profit margins. And so a lot of the businesses moved international. There aren't that many big VFX houses in the United States. And so it's all just a very complicated business. So Warner Brothers Discovery has actually filed a suit against Midjourney and Disney and Comcast. Universal also has sued Midjourney for making copies of their copyrighted properties. The script for Critters was written by some of the members of the team that wrote Paddington in Peru.

B (40:07)

Interesting, though. So making copies of their copyrighted properties. This maybe sounds different than the anthropic lawsuit with the writers where it was like, anthropic Was using Libgen.

A (40:19)

Yeah.

B (40:20)

And not paying in this case. This might be that Mid Journey actually bought the videos, but then made copies of them.

A (40:26)

Interesting.

B (40:26)

During the training process. But I don't know. I don't know how much we should read into that.

A (40:30)

I mean, these, at least. The Warner Brothers discovery lawsuit just filed last week, so I imagine that they haven't even gone through discovery. No pun intended.

B (40:41)

But if lawyers do lose their jobs to AI, they are certainly going to cash out on the way out on these copyright lawsuits. The script for Critters was written by some members of the team that wrote Paddington in Peru. And again, John, your big movie.

A (40:58)

Go.

B (40:58)

Did you see?

A (40:58)

I saw Paddington in Peru.

B (41:00)

Really?

A (41:00)

It wasn't as good as the first or second Paddington's. I would put the first.

B (41:04)

Is it a. Is it an adult movie or.

A (41:06)

No, it's a kids movie with my son.

B (41:08)

Yeah, it's fantastic. You made it sound like you didn't just watch it, you studied it.

A (41:12)

Oh, I. I have sat myself down and watched Paddington on multiple occasions. It's one of the greatest films of all time.

B (41:19)

Okay.

A (41:19)

Paddington is actually. I think it's at the top of the IMDb ratings. It's like one of the greatest movies of.

B (41:24)

No way. Oh, so out of the loop.

A (41:26)

No, Paddington is incredible. I highly, highly recommend Paddington. It's a great movie. Anyway, I get.

B (41:33)

I get on it.

A (41:34)

OpenAI is betting that if Critters is successful, it will show that I can deliver strong content, strong enough for the big screen and accelerate Hollywood's adoption of the technology. Nelson said OpenAI's tools can also lower the cost of entry, allowing more people to make creative content. Now, the question this is.

B (41:50)

I mean, on this note, what I've been saying is, is right now there's sort of a certain amount of budget every year provided by various groups to fund films. And if you just reduce the cost, even by, you know, 60, 70%, like, well, if you look, we could potentially just get way, way, way more films. Right? The same. And potentially it doesn't necessarily. I mean, it doesn't necessarily suck out the profits too, in the industry, Right. Because, yeah, it's just potentially a budget of $100 million. It would have gone to one film, now goes to three different groups that are each making their own film.

A (42:23)

No.

B (42:23)

So it could be. Could end up being a win.

A (42:27)

Yeah. I mean, I think the dynamic here is that it is extremely telling that they are still using humans to come up with the concept and come up the script and do the voice acting and it's not an attempt at one shotting an entire film using AI from start to finish. The prompt is not make a movie for $30 million that makes more than $30 million. Like that is the prompt that you give a studio executive, like when you go that is the job of, of a real studio executive is just make.

B (42:59)

Money, don't make mistakes.

A (43:00)

Don't make mistakes, right? And then they have to go out and find a script and they have to find an artist and they have to find a post production house and distribution and marketing campaign and they have to make a bunch of really great entrepreneurial decisions to actually return the capital that's invested. Now with this, I see AI as being used as a tool, much like the original Pixar Renderman was used as a tool to create Toy story or Cinema 4D or Houdini are used to generate hordes of battles during Game of Thrones, et cetera. And so this should wind up being just another option in the tool chest.

B (43:40)

For Goldrock is also saying you take that same budget and if less has to go into production then you can put more into marketing and events surrounding it, activations, et cetera.

A (43:51)

I mean, I would imagine that the long term effect of this should be something similar to what cloud computing did for startups. So before it was like raise $10 million and make sure that you have not a data center but like you have a closet with racks of servers because if you want to build a website you need to buy a server. And then it became sign up for the free tier of AWS or Google. And so what wound up happening? Well, we got a lot more startups, but that didn't mean that everyone made money. There was a lot, there were a ton of failures, but we got a lot more shots on goal and because the bar was lower, we got a very steep power law outcome, but we get more value overall. And I would imagine that that's what happens with movies is that anyone can create a movie doesn't mean that everyone's going to create a movie that makes money or is great or loved. There's going to be a lot of slop that people don't like at all. Just like think about all the startups that are out there that don't do it, don't do well, well, but they're.

B (44:49)

Able to when you, when you watch a, a regular movie. I've done this a couple times. I think you've done it quite a bit, quite a bit more. And it's bad. Yeah, it really is. You guys Went to all this effort.

A (45:02)

Totally.

B (45:02)

All, all these months, all, you know, shooting all this content to make a bad movie.

A (45:07)

And how many times is that because, oh, oh, the camera wasn't working or something. Or like.

B (45:11)

Yeah, oh, no, it's usually the VFX delivery.

A (45:14)

Exactly. It's like, it's the, it's the art of bringing everything together. And that's something that I think is going to stick in human world for a long time. Yeah. Because. Yeah. I mean there's a bunch of reasons of how complex it is, how hard it is to build reinforcement learning on top of it. It's very hard to build this like reinforcement learning environment around just make a profitable film like that takes years. It's a huge investment. It's very hard to reverse engineer. Anyway, let me tell you about Vanta Automate Compliance. Manage Risk and Prove trust continuously. Vanta's trust management platform takes the manual work out of your security and components compliance process and replaces it with continuous automation. Whether you're pursuing your first framework or managing a complex.

B (45:55)

And now would be a great time to talk about Fall. Yes, the generative media platform for developers. We've had Gorkham, the CEO, on the show multiple times. This company is growing at a absolutely ridiculous rate. It just crossed a $100 million run rate, I believe it was a week or two ago. And Fall works with a bunch of different companies. Adobe Canva. What everyone wants this Eagle sound effect, the Eagle effect.

A (46:24)

Apparently that's for Turbo Puffer, but I.

B (46:26)

Think we can use it for everyone. It's just so good. But anyways, if you want to leverage all of these new image, video, audio models through a single API, you can go over to Fall AI and check it out. And happy to introduce you to the team if you're interested as well. But this company is an absolute monster and we're pumped to be partnering with them.

A (46:50)

We're very excited. That's our latest, our latest sponsor of the show. They make it possible. Anyway, let's go through some timeline. Lululemon is now the worst performer of the S&P 500 year to date. I didn't even realize they were in the S&P 500. That's. I mean, that is an accomplishment.

B (47:08)

They're in the league.

A (47:09)

The Stock is down 56%. Oh, the Trade Desk is down 55%. The Trade Desk was a company that people were really praising for. Just like elegant and incredible. There was some narrative violation around the trade desk. I need to dig into that. But it was something like they hadn't raised any money or something like that was why they were so.

B (47:34)

Yeah, I mean, it's interesting that Nike.

A (47:37)

Gartner is down 50%.

B (47:40)

Gartner's in the trough. Gardner is in the trough. You created Gartner.

A (47:47)

Gartner.

B (47:48)

All right, let's keep it together. What comes after the trough of disillusionment? The plateau of productivity.

A (47:54)

No, no, it's not the plateau of productivity after the trough of disillusionment is the slope of enlightenment.

B (48:00)

Okay, okay, okay. And so don't worry saying the slope of enlightenment, it's ahead. There's, there's, it's only up from here.

A (48:07)

It's only up from here.

B (48:08)

If the hype cycle holds true. It's only up from here. But these are some. It's interesting. Nike and Lululemon, you know, both. Nike peaked in, in 2021. They obviously went heavy into E commerce. And Lululemon, you know, I don't, I don't necessarily think they always had a pretty big E comm presence, but I think this is a case of just getting eaten alive by Alo and Fori and all these sort of new entrants that are going after that same market of like, kind of non athlete sort of wellness. Right. And yeah, this, this tells me they just basically like took the market for athleisure and just divided up amongst themselves.

A (48:56)

Yep. People are also dunking on them for not reading the room and kind of staying in that like, Jaguar rebrand world and not pivoting fast enough to the new Sydney Sweeney American Eagle style campaign of Americana. And so I think if they want to bring it back, there's one easy solution. They should hire Lulu. Do you start going direct? I think that's the move.

B (49:24)

Something there, something there.

A (49:26)

They should also get. I mean, wait.dev, code review for the Age of AI graphite helps teams on GitHub ship higher quality software faster. Update that website Lululemon. Get on graphite.

B (49:37)

Well, you know, you remember Chip used to go direct and that didn't go so well. Or Chip Wilson, founder of Lululemon.

A (49:47)

Oh, I didn't know that.

B (49:47)

No, he did a. He had a really famous interview where people were saying, like, you know, Lululemon doesn't fit everyone. And he said, like, my clothes aren't for everyone. Basically implying that, like, he didn't make clothing for people that weren't fit.

A (50:03)

Oh, interesting. And then there was a backlash to that and then they became kind of, you know, more inclusive. Is that the idea of like the cycle? Because now they're getting. They're getting dragged at least in the crazy X comments about. About being like, too inclusive or something like that. Not Sydney Sweeney, American Eagle Coded enough. Anyway, that's wild.

B (50:25)

Yeah. Founder says brand is not for everybody.

A (50:29)

You know what's funny is that there's a fantastic video of the. It's not the founder of Lamborghini, but it's like the head of sales of Lamborghini when the Lamborghini Countach came out in the 80s and he's doing a TV interview, and the interviewer says, who is this car for? Is this car for everyone? And he says, no, this car is not for everyone. This Lamborghini Countach is a V12, right? Is it a V12?

B (50:56)

I think so.

A (50:57)

And it's incredibly tight. Doug Demuro, I think he can't drive it with his shoes. He has to take his shoes off because it doesn't fit. And it's this insanely loud, insanely engaging, crazy experience. He says this is for someone who has an extra car that travels behind them with their luggage. And he says some other wild stuff about, like, who the Lamborghini Countach buyer is. And so truly, I don't think that there's anything necessarily wrong with having a small market or defining the market and excluding people from the market. Like Lamborghini Countach. Like the Countach. It's not for everyone. I think that's fine. I think they can make it more desirable, but you have to price accordingly and you have to actually deliver at that level. And so Lululemon might not have ever been the Countach of clothing.

B (51:47)

Yeah, it's different. This is before GLP1s, right. This is Lululemon, theoretically would have a buyer that is buying the products to get fit and get healthy. And so for the founder to be like, you know, it's different too. I think they were a public company by that time already kind of. Kind of a credible story of the company overall, though.

A (52:09)

Anyway, good luck to them. Hopefully they build back. Hopefully they ride the hype cycle alongside Gartner into the plateau of productivity.

B (52:19)

Pretty. Pretty funny. There's a video here of Putin's advisor, Kobiakov went out on record and saying the US is going to shove debt into stablecoins and to devalue it.

A (52:36)

Russia just accused of the US of using crypto to wipe out its $35 trillion debt. And Rex is, that's exactly what we're going to do. Yeah, they're going to socialize.

B (52:47)

Yeah, socialize. Good. Alexander went pretty viral. I think it was on Friday or Saturday going on this like 20 minute rant talking about stablecoins and gambling and hyper capitalism. And yet obviously the dollar is a way to export our debt. And stablecoins can potentially accelerate that. It's potentially a situation where it's sort of like net positive for the world. If people can access a more stable currency, they can participate in our.

A (53:26)

It's less positive than it is right now because it's basically like you've been getting global stability and open trade routes sort of for free without having to participate in that debt that pays for it. No, I mean you will also be.

B (53:42)

Paying by nature of needing to transact and settle trade and dollars. You already were financing. Financing.

A (53:50)

Yeah, yeah, but on the margin more people will be transacting in dollars and therefore more people will be. So will be socializing and holding the.

B (53:56)

Dollar because it's, you know, I mean, and this is a question we've brought this up with a number of people when talking about stablecoins is at what point does foreign government just say like, no, we don't want stable. You know, we don't want our citizens to hold stables.

A (54:11)

Right.

B (54:11)

We want you to hold our money.

A (54:14)

Yeah, it seems like a crazy, crazy move to be like, yeah, I'll just give up my own currency and dollarize. But we've been in the game of dollarizing for a while. We like dollarization. It's not our first, not our first dollarization.

B (54:25)

Don't go a dollar with us. This is interesting. Unitree Robotics is filing for an IPO at a $7 billion valuation. Annual revenue of 140 million. Way, way less than thought. Yeah, they have 65% of the robot. 65% of the revenue is from robot dogs. Let's give it up for robot dogs.

A (54:48)

And that's 70% of the global market.

B (54:51)

Who's got the other 30%?

A (54:54)

So 65%. They must be doing a hundred million dollars of robot dogs. And the total market of robot dogs globally must be like 130 million.

B (55:01)

I guess the dogs are barking, John.

A (55:05)

But that's like not a lot of.

B (55:06)

I mean, still, still the main, main thing. This going out at 7 billion is not good for a certain paying 50x revenue. Well, well, I'm just saying, I'm not.

A (55:20)

Saying, but I'm just saying even though.

B (55:23)

There'S people out there in the, the humanoid space that don't want comps out there.

A (55:27)

Yes, yes, yes.

B (55:27)

No comps. Please keep the, keep the, keep the comps out of the market.

A (55:33)

Yes. Yeah, but is this overall, where is this going to ipo? Is this going to IPO in America? That would be crazy. But who knows? So while he looks that up, 30% is from the humanoid robot. 5% is from the sales of sensors, actuators and controllers. It's happening, says Nick. You love to see it.

B (55:53)

I love how you can take any news and make it seem like it's so over or so back. The IPO is going to be on a Chinese stock exchange.

A (56:05)

Well, if you want to trade on the American stock exchanges like American patriot, go to public.com investing for those who take it seriously. They got industry leading yield, industry leading yields. They're multi asset investing and they're trusted by millions.

B (56:21)

We hung out with earlier today's team this morning. They made us a cake.

A (56:27)

It's very nice of them.

B (56:28)

Said thank you, congrats for nothing. Get back to work.

A (56:31)

It was good. It was very funny. And it was revealed in this hilarious way of like showing up to breakfast with this massive box that like clearly has a cake in it. But he's like, I'm not going to tell you what's in here. And I'm like, that's obviously a cake dude. But he's like, no, no, no, I gotta reveal it once you're all sitting down. So he sit down and he opens.

B (56:48)

It up like congratulations.

A (56:49)

But then I was surprised because the joke on top was very funny.

B (56:52)

Yeah, the joke on.

A (56:54)

Anyway, Elizabeth Holmes has been on an absolute terror poster in residence poster in prison. So Elizabeth Holmes posted some. What would I tell the 19 year old girl who was getting ready to drop out of Stanford to build Theranos, doing some thought leadership posting. And Gonto says, since when are we taking advice from somebody who went to jail? There should be a community note in the post. And Elizabeth Holmes says, let's listen to a growth Exec who has 8k followers. Brutal ratioed. Rohit says, beginning to like Elizabeth Holmes. Very funny. I don't, I don't have a problem with taking advice from people who went to jail. They probably learned a lot. I believe in restorative justice, although she is still in jail.

B (57:38)

So yeah, it's, I think this is gonna be the new playbook of like when a public figure turns the entire world against them. Wait, hire a ghostwriter and then hire ghostwriter. Start posting.

A (57:52)

Start posting for sure. But I mean you have to be, you have to be huge. Like Elizabeth Holmes has a movie and a documentary and a book.

B (58:01)

Shkreli did this.

A (58:02)

Yeah, but Shkreli didn't even get the treatment that Holmes Did. I would say that Elizabeth Holmes is more of a household name than Shkreli.

B (58:09)

More, I think. Yeah, she is, but. But he's also. I mean, he's just so good on the mic. Yeah, definitely got people.

A (58:16)

I actually. I DM'd Elizabeth Holmes, invited her on the show via a phone interview.

B (58:21)

Are we sure that it's actually her?

A (58:22)

No, we're not. It's probably her husband, but.

B (58:26)

Well, yeah, it's her handwriting. Every post, maybe. No. Yeah, I mean, just reacting too quickly.

A (58:32)

I mean, there's probably a little bit of her here, for sure. But I do think it's mostly the husband who's on the outside. But I can't confirm that that's pure conspiracy theory. But we remember when we were talking about Elizabeth Holmes, like, months ago, and I was saying, I'm ready to hear her out, but I want her to talk about biology, specifically, I want her to. That is the path to redemption is give me some banger takes about what's going on in GLP1s, what's going on in MRNA, what's going on in DNA analysis and sequencing. Like, show me that you are truly generational, on the cutting edge. Because I feel like Martin Shkreli's been doing that. He was like, I was at a hedge fund. I went to jail. Now I'm coming out. What am I giving you? I'm giving you hedge fund, like, takes and my takes. A lot of people agree that they're pretty good. And so Shkreli has somewhat redeemed himself in the sense that, like, he might have done something wrong and wound up in prison and gotten out and served his time. But the key thing is that his initial claim of, like, I'm a finance guy is sort of continuing to ring true. But none of Elizabeth Holmes posts thus far scream like, wow, really differentiated view in bio. Like, there is alpha in understanding the bio world. If I listen to her, like, yeah, she's in prison, but clearly she's a great biologist, clearly she's a great scientist. She messed up on the finance side, defrauded investors, went to jail. But I need to listen to her because she understands where the puck is going in terms of biology or science, and I'm not seeing that yet. So I am. I'm. I'm withholding my. My endorsement of her anyway. If you want to be on the cutting edge of science or data analysis, go to Julius. What analysis do you want to run? Chat with your data and get expert level insights in seconds. They're loved by over 2 million users.

B (60:35)

And trusted individuals in Business Insider this morning. Looking like an absolute chat though the.

A (60:41)

Photos they sent, great. They sent a great photographer. Really, really good. Dart says, I moved to a play a new place two years ago. High risk area, job related, not exactly by choice. So I made a bunch of these signs and put them up everywhere. Two years in and my house has pretty much been the only one in the entire neighborhood that hasn't been broken into. And it says protected by Palantir home security. That of course is fake, but you know, he put it up there. I think this whole thing's just a funny post. I don't think that the average home intruder knows what Palantir is, but still, very funny post.

B (61:18)

I don't know, maybe. Maybe they're part of the retail army.

A (61:22)

Maybe they're like, oh, oh. For that reason I'm not invading.

B (61:26)

Maybe they're breaking more Palantir shares because they're so bullish.

A (61:32)

Oh, they're breaking into other houses. But then they see that as a sign of respect.

B (61:35)

Yeah.

A (61:36)

Say I'm so long Palantir that I won't break into this house.

B (61:38)

Yeah. They just believe in the company.

A (61:40)

That makes sense. Totally possible.

B (61:42)

Kylie Robinson hit the timeline. She fired back with a quote tweet and she says listening to this lol. I did not go into the review expecting to hate it. Also was aware I'm not the target audience for an always listening device. And she highlights, she puts us in the true sound highlights a part of the article. She says, I'll admit I'm not the target audience. I imagine the person who'd want a friend is someone who is likely not a journalist who may have more social occasions where they can sport an always listening pendant. And anyways, so I feel like journalists.

A (62:16)

Should be perfect for an always on listening device because every time you talk.

B (62:19)

To them on you're like, do you.

A (62:21)

Mind if I record this? They're like, let me pull out my phone and record this. You should just have an always on recording device is like the perfect device for the journalist. It's in fact like the new notebook for the journalist.

B (62:33)

Yeah. And you could just say let me know if you ever want to go off the record.

A (62:36)

Exactly. By default you're on the record and I'm recording. I think that, I think that, I.

B (62:41)

Don'T know, I think long term we.

A (62:43)

Might see massive product market fit with frame. They should do enterprise. Avi should do an enterprise product for B2B products.

B (62:51)

Yeah.

A (62:51)

Yeah. So the journalist gets it, wears it and as soon as you Interact with the journalists. You know that you're being recorded and everything you say is on the record. Speaking of journalists, Julie Hornstein has a little deep dive on Raul and some other young founders. They've long been icons, she says. She wrote about how because of AI hype and ease of vibe coding, teens and 20 somethings are flooding silicon Valley to build startups. Instead of attending college or securing big tech jobs, founders told me that they'd rather start a company than go to college because AI gives you a PhD in your pocket. As Jake Adler said. Love Jake. Because of mass layoffs, Roy Lee thinks the worst thing someone who wants to be, who wants a stable future could do is work in big tech. Interesting. Roy Lee coming in with a hot take, as he's known to do college is dead. Says zero interest rates. He graduated from university in 2019. Marvin von Hagen, 25. I feel like I was just texting with Marvin. Is that right? Marvin? I think I gotta text him back. No, literally, I do. Yeah, he texted me. He said, hey, we got an intro. He's releasing a short film. We gotta play this poke. Yeah, we gotta get him on the show. Anyway, 30% of the Y Combinator batch were college students or new grads. So nice little trend piece with some beautiful, beautiful photography. So shout out to everyone that was featured in Silicon Valley's Youthquake. In the age of AI, founders aren't waiting to grow up, says Business Insider.

B (64:28)

Will Manitis says basically everyone underestimates how big of a role office selection plays in the outcomes of companies and their cultures. I recently had the chance to spend 24 hours at Lego headquarters. Will you enter and then stay the night?

A (64:44)

Did you not get the memo about the great locket? You're not just supposed to be going.

B (64:48)

And hanging out at the Lego LEGO.

A (64:50)

Headquarters going, kid mode. This is a dream come true for any child. Yeah, but, you know, that's what they say about Will.

B (64:58)

No, but I'm just. I'm still kind of hung up. 24 hours at Lego headquarters. Did he enter at like one bait?

A (65:05)

It's so good. It's the best bait. Bait.

B (65:08)

But I do. Like, you can see they have, they have that Lego. On the actual side of the building.

A (65:12)

We have a fish. I feel like we had a fish at some point. For bait. Yes. So LEGO is a relatively. Lego is a relatively strange company for the toy industry. Even at $10 billion in annualized revenue, the company has never taken a dollar of equity. Financing is still family owned and controlled. Since 1932, the company is based in Belund, Denmark, a farm town of just 7,000. And where they're founded is still where they are. That's pretty remarkable. The company and the Christensen family through their holding company Kirkbi, functionally own the town of Belund. From the airport majority control to a large majority of the real estate and the hotel tourist infrastructure. This allows them to make remarkably long term decisions like 24 hour visits for random American entrepreneurs.

B (66:03)

Let's pull an all nighter together.

A (66:04)

We're thinking long term.

B (66:05)

No, this is.

A (66:06)

Maybe you want to pick my brain. Let's think long term.

B (66:10)

The Lego execs are participating in the great lock in. Maybe said come visit but you got to be locked in with us 24 hours.

A (66:17)

Yes.

B (66:17)

Pull an all nighter.

A (66:18)

We got to get more info here.

B (66:20)

Were you a big Legoland kid growing up?

A (66:22)

Not a big Legoland kid. Legoland can date myself built like I remember when it was built and I remember when it like opened and I was like there on like one of the first days but only went like once or twice. But very cool support Legoland. Big, big into Legos generally.

B (66:37)

Yeah, huge, huge stepping on them.

A (66:40)

Yes. One of the worst pains imagined. While some of this investment has been company related, much of it has not been. The family has been actively involved in turning a blighted racetrack into a new neighborhood and improving, improving civic and school infrastructure. This kind of long term investment in a place is only really possible with a large balance sheet and patient private control. This was a rare thing before the last 10 years, but many tech companies could now fit this mold. That's interesting. My prediction is we will see many more company towns emerge. So I mean this has kind of happened. I feel like Menlo park is very nice and Cupertino is very nice. And then a lot of that's just a benefit of, you know, Apple sets up a massive headquarters there. All the employees are wealthy and so the town, the taxes.

B (67:28)

Yeah.

A (67:28)

The foils. Yeah.

B (67:29)

The California Forever project could turn into this. Right. If you get a single tenant in there that's doing like shipbuilding.

A (67:36)

Yeah. But you need a company to be the backbone of the flywheel and that's certainly what's going on in Denmark. The results of this investment are clear. Even on a Saturday, many employees were walking, biking to the office with families in tow. The town felt vibrant with a mix of visitors, employees and locals. So now we're finding out that he spent 24 hours there on a Saturday, but he'd get there Friday night and didn't leave until Saturday night or he slept there on Saturday night and woke up on Sunday and left. I'm so confused. So he says it seems like a remarkably pleasant place to live and work. That eliminates many of the trade offs of highly urban office settings. As we see the emergence of dozen late stage privates with fortress balance sheets, it's hard to imagine the company town not returning. I like it a lot. It's a good time.

B (68:26)

I would like to do a TV company town.

A (68:29)

I would love that. I would also love a turbo Puffer company town. Plain search everybyte, serverless vector and full text search from first principles on object storage. Fast 10x cheaper and extremely scalable. Get started for free.

B (68:44)

Be like linear, be like cursor. Head over to turbopuffer Ocean. It's time to puff.

A (68:49)

So Scott Bessant, he is on a roll. He's got a little bit of a little bit of a penchant for curse words.

B (68:59)

But I don't normally like curse words. I kind of like when he does it.

A (69:03)

Yeah, but it won Geiger Capital over. Geiger Capital says I'd follow Scott Bassett into war. And Scott Bessant said why are. Why the f are you talking to the President about me? F you, I'll punch you in your effing face. And Geiger Capital says that's my treasury secretary. Anyway, just a funny quote. I don't know how that got into Try this line.

B (69:28)

If you have a co worker, you heard them talking, talking to management in a not so kind way about you. Try this out. Check this out in the. You know, around the water cooler.

A (69:40)

Keep the President from talking about you. Keep chatgpt talking about you. Go to profound. Get your brand mentioned in ChatGPT. Reach millions of new consumers who are using AI to discover new products and.

B (69:53)

Brands and Elike MongoDB indeed. Mercury ramp Zapier workable US bank and.

A (70:01)

I saw a fun post in the timeline. Somebody asked what's the best way to get your brand mentioned in ChatGPT? And the answer was profound. So clearly profound has been dog food.

B (70:10)

They're dog fooding.

A (70:12)

So Preview of what might happen at Meta Connect Jukan has a screenshot here from Semicon Sam I was considering releasing it as a paid post, but after much thought I decided to make it free. I hope you all enjoy my analysis on smart glasses. Why smart glasses are already dominated by China. Why even Apple and Meta have no choice but to cooperate. So this article says. Recently there was an article like this. Meta will unveil new smart glasses called Hypernova at its annual connect event on September 17th and 18th. The prevailing view was that unlike the previously released Ray Ban metaglasses, the Hypernova coming this time would have a display and its price would be high above $1400. However, contrary to expectations, according to a report by Mark Gurman, Absolute Dog, it will launch at $800. So Ghermin kind of runs through it. It's still higher than the Meta Ray Bans that are around $200 to $400, but should be a little should be still pricey, but lower than people initially expected. So the question trying to be answered by this Substack article is why is that? I wanted to find out the reason. I want to find the reason in competition with Chinese manufacturers About a month ago Alibaba released smart glasses called Quark Vision. But the price is 1,999 yuan, about 280 bucks.

B (71:37)

That's pretty cheap. I'm not laughing at the price, I'm laughing at the name quarkvision. You ever want Quark?

A (71:42)

Quarkvision?

B (71:43)

It's time for Quarkvision.

A (71:46)

I don't know, maybe it sounds better in Chinese. So from Meta's perspective, with China putting products out that cheaply, they might have thought how can we manage to charge more than $1,400? And decided to set a lower. I believe smart glasses will become a truly essential device. This writer is Smart Glasses pilled I'm also optimistic about the growth in smart glasses generally VR, Arkansas I think we're about to see the churn rates drop and the adoption rates increase. Let me give an example. Unlike smartphones, smart glasses are much more convenient to operate. You don't need to move your fingers. You can operate them through humanity's most concise action, your voice. For instance, if you're trying to use AI features like Gemini Live on a smartphone, you would have to open the camera app with your finger scan things yourself with AI glasses Camera on the device is effectively seeing what I'm seeing. So there's the advantage that I don't have to issue instructions one by one. And so he lays out kind of a bull case for the smart glasses. This is the industry's classification of smart glasses by generation. Gen 1, which is the current generation, doesn't have a display. AI features are supported through a smartphone integration. Gen 2, which is next year, supposedly will be equipped with a low resolution display and then Gen 3 has a high end display, full color 2K resolution AI and spatial computing like what we saw in the Apple Vision Pro. So starting from so he says that's why I believe the second generation is the point where we can start calling them smart glasses. From the second generation onward, displays provide visual information, allowing users to access much richer data through the glasses. To draw an analogy, up to the first generation they were like PDAs, but starting from the second generation, they can be called smartphones. So very bullish on the growth here, and I think we can Was there anything else in here that you wanted to run through?

B (73:42)

Yeah, I just thought the pricing information was interesting. Yeah, they could try to run the same playbook of just selling the same playbook that they ran with like dji. Right. Of just like selling below the actual cost to produce the devices to just try to get adoption. Yeah, but we'll see.

A (74:03)

It is interesting. I haven't, I haven't seen that much reporting on the bill of materials for the Oculus Quest 3, but I don't believe that they were selling it at a massive discount. I do think it was probably profitable on a per unit basis, but then they were just spending so much on R and D to actually build the next versions. It would be very interesting to see what happens if they released something at the level of Apple Vision Pro, but at the price point of the quest 4 or something like that. I really, I'm excited about the smart glasses, but I'm honestly more excited about just an Apple Vision Pro level image and screen being pulled into something with the Quest's ergonomics.

B (74:49)

So there's some interesting. So basically, brightness and power efficiency are the most important things when it comes to displays. And the reason that Apple's Vision Pro can crush it on brightness is because the field of view is actually blocked, so there's no outside light coming in. And so they can deliver this like ultra crisp image. And what Meta and Quark are doing is like you can just see the real world and see a display on it. So the sort of quality of the image will lag behind that. Just like ultra high fidelity that you're getting with the Vision Pro. Yeah, I mean, obviously insanely real trade off.

A (75:26)

I mean, the physics of light state that, but it's an additive process. So if you're outside and it's bright and you want to project a black cube into that world, that's basically impossible from a physics perspective because the bright light's going to shine through, you're adding light on top of it, and there's not really anything that you can do to make the black cube show up on the beach in Santa Monica. If you're just walking Around. So there are some serious headwinds there. That's why Palmer Luckey predicted that the end result of augmented reality would be reprojection, which is what the Apple Vision Pro does. Take a camera. Because then if you take a camera and you play that on the inside of the display, you can turn down the brightness all around the black cube, and then you can put the black cube over there and you can block out the light that's behind it. But occlusion with dark objects in augmented reality scenarios is basically physically impossible. I don't know. Seems very difficult. Anyway, whatever. If you're planning to launch a augmented reality headset, you gotta get on linear. LINEAR is a purpose built tool for planning and building products. Meet the system for modern software development, streamline issues, projects and product roadmaps. JIRA ticket says Houston is like if the AWS console.

B (76:45)

Got it. Tyler there. What's going on?

A (76:46)

Yeah, how are you doing, Tyler? What's new? I was just clapping.

C (76:48)

I was just so excited about the linear.

A (76:51)

Good.

B (76:52)

Up to you.

A (76:52)

Tyler, what is burning up the timeline in your world? What was your favorite post from this weekend? Okay, let me find a really good one. I'll get back. Okay. In the meantime, let's talk about Houston, the city in Texas. Houston is like if the AWS console was a city, says JIRA tickets. And Kathleen Turner says, dang, why? And JIRA tickets says, I don't know. I've never been to Houston. Like, I don't even know what this means. I don't know why this wound up in the show or in the run of show. I think this wound up. It wound up in the run of show, like multiple times for some reason.

B (77:26)

Nice.

A (77:27)

But did you put it in?

B (77:28)

I put it in. I thought it was funny.

A (77:29)

Why? What appealed to you about this? Do you think it's an apt analogy? Have you ever used the AWS console? Do you know how confusing it is?

B (77:37)

Yes.

A (77:37)

Okay.

B (77:37)

Yes.

A (77:38)

So is that the take? It's just saying it's really confusing.

B (77:41)

Best designed city.

A (77:42)

Okay, yeah, I get that.

B (77:44)

But like, it's also just funny.

A (77:45)

A lot of horsepower.

B (77:46)

It's a good post, sir.

A (77:47)

It's a good post. It's a good post. Shout out JIRA tickets. Asml.

B (77:51)

Yeah, this was. This was cool. ASML decided to become a VC according to Wasteland Capital to boost European tech sovereignty and burn one and a half billion in shareholder funds into Mistral.

A (78:03)

Hey, we don't know that they burned it.

B (78:06)

They yoloed it in lithography. BUN plus third tier LLM meat with a topping of EU bureaucrat diriges deridges me. I don't know that word that I can't pronounce and you can't pronounce meaning that this might be a new type of AI hamburger. Anyways, people are not control of economic matters. So people are not excited about ASML's investment. They did about 3/4 of Mistral $2 billion Series C Mistral is getting valued at just north of Cognition, its latest round. And anyways, people are basically saying ASML was potentially worried about being too much of a private company and just focusing on generating profits and they have to redistribute some of that wealth to the broader market.

A (79:05)

Indeed.

B (79:07)

But anyways, it might be, I mean if Mistral, I think clearly it's France's national champion and effectively one of Europe's champions and I'm sure they can figure out a way to create value and hopefully ASML shareholders get a nice Doug.

A (79:29)

O' Laughlin over fabricated knowledge said, this is very simple. National champion. Don't overthink it. There isn't a 4D chess move. This is checkers. So I actually don't know exactly what that, like what the interpretation of that thesis is like. National champion just means like you must be supported at all costs. Right. And so like it doesn't make, it doesn't necessarily matter if it pencils out on, on some DCF right now. It's like we need to continue to support that. We have the capital, so let's continue to support our national champion. Makes sense. You want to talk about EchoStar?

B (80:02)

Yeah. So SpaceX is buying EchoStar's AWS 4 and H block spectrum licenses for about 17 billion split. It's up to 8.5 billion in cash and 8 1/2 billion in SpaceX stock, which is crazy. The big takeaway, the deal allows SpaceX to expand its direct to device mobile services more independently beyond its existing T Mobile partnership. I'm going to pull this up. EchoStar. So this is an industry that's heavily regulated and you would think that in a more free market SpaceX could just say, we're going to start building.

A (80:48)

We'Re.

B (80:48)

Going to just create this like mobile Internet service. And it doesn't work like that.

A (80:53)

There are specific licenses. You don't want traffic on the same spectrum. Band of the spectrum, yeah.

B (81:00)

Shares of EchoStar rose as much as 26% on Monday to a record high of $84. Its bonds were the biggest gainers in the junk bond market. Yeah, apparently EchoStar was like potentially veering towards bankruptcy. Like they hadn't been.

A (81:15)

Wait, so what did EchoStar's stock do? Up 26% because now, now it's basically a SpaceX holdco. Like because they've 8.5 billion in SpaceX stock. What's their market cap?

B (81:26)

So it should trade at like 10 trillion potentially.

A (81:30)

What's their market cap?

B (81:31)

22 billion.

A (81:33)

Whoa. Wow. This is.

B (81:35)

We were just talking, we were talking about this. So we were talking about this morning. There's all these crypto treasuries.

A (81:40)

Yep.

B (81:40)

And then you would think, like, why doesn't somebody create like the Elon Musk, like Treasury company? And the reason for that is like if you start adding shares, then you become a registered investment advisor and there's all this compliance. Much harder than just putting like digital assets or tokens in a company.

A (81:55)

Yeah. These are just like I have USD, I have bitcoin. They're kind of the same thing. I'm putting my treasury wherever. And then Microstrategy now just strategy wound up just having a ton of Bitcoin and then it became this, like this way to invest in bitcoin just with a public ticker. But EchoStar will now be the most concentrated way to get, to get allocation into SpaceX. Right.

B (82:17)

So crazy.

A (82:17)

That is crazy.

B (82:18)

If you bought it Friday.

A (82:20)

Yeah.

B (82:20)

The, basically the value of the business then was just like SpaceX.

A (82:26)

Yeah.

B (82:27)

Plus the cash they got from this deal, like the SpaceX stock.

A (82:30)

Yeah.

B (82:31)

So the company is sitting on this.

A (82:33)

License and they just sold it, flipped it into SpaceX stock stock.

B (82:37)

Crazy.

A (82:38)

That is a very.

B (82:39)

I wonder, I wonder if they'd.

A (82:41)

Ludlow. The bigger takeaway is I think we got a SpaceX meme stock on our hands soon. This is crazy.

B (82:47)

Oh, you're saying this is your take?

A (82:49)

Well, yeah, yeah. So Ed Ludlow, his, his big takeaway, which I agree with, I think you are correct. The big takeaway is the deal allows SpaceX to expand its direct to device mobile services more independently beyond needing to partner with T Mobile, they can go direct and cut out T Mobile. Extremely bullish for SpaceX direct to device. Obviously that's going to be a huge business and they're really making a lot of inroads there with partnerships and now owning the actual license. But the other big takeaway is that now there is a company that, where, what, 25% of their market cap is directly indexed to SpaceX and it's just a public ticker. That is a crazy development cash.

B (83:29)

They're also burning money. They lost 300 million last quarter.

A (83:35)

Hopefully they'll make it all back on Elon Co. Who knows? Anyway, Numeral hq.

B (83:44)

The only other thing.

A (83:46)

Yeah, you go for the sales tax on autopilot. Spend less than five minutes per month on sales tax compliance. While you're looking that up, Tyler, what did you find for me?

B (83:54)

All right, so I think my favorite.

C (83:56)

Post was Will Brown.

A (83:58)

Oh yeah. He said he was at a family wedding in Spain. This is brought a printed out semi analysis cluster max list of the, you know, the neoclouds.

B (84:08)

He was asking why Prime Intellect is at the bottom. Hey, made the list.

A (84:14)

He made the list. That's great. Gotta start earning the lead at a family wedding in Spain. Uncle brought a print out of the semianalysis cluster mass list and asked why we're underperforming. Explaining the difference between marketplaces and data centers and how we're not just selling compute, we're also advancing RL infra. I love it. Very fun.

B (84:33)

So one small problem with the meme stock potential Echo stock.

A (84:39)

Break it down.

B (84:41)

They've got 29 billion of debt that they had been basically defaulting on.

A (84:49)

Maybe they'll be selling the SpaceX stock off slowly to pay for the debt.

B (84:53)

Honestly, just hold, just hold. Just make the minimum payments on the interest and just hold on and just hope Elon gets tomorrow.

A (85:00)

So you're telling me that the narrative around this company is that it's a levered bet on SpaceX in the public markets that are the ticker that anyone can trade.

B (85:10)

Yes.

A (85:11)

And the company's already levered up.

B (85:12)

Yes. You can think.

A (85:13)

They don't, they're not. They don't even need to lever up further. Yeah, they're already levered.

B (85:17)

Yeah, yeah, yeah.

A (85:18)

Wow.

B (85:18)

Extremely levered. Except unfortunately they, they didn't raise debt to buy SpaceX. They just.

A (85:26)

They already had that.

B (85:27)

They already had it because they already have. They don't need.

A (85:30)

Oh, there are two steps. Most people are like, I'm going to become a Treasury company, then I'll lever up to buy more bitcoin and my company will act as like a 2x levered Bitcoin. This company's already levered up.

B (85:43)

So basically SpaceX just needs to. SpaceX just needs to be a 1.7 trillion dollar company and then they'll be able to just pay off the debt.

A (85:56)

There we go.

B (85:57)

And pay off the debt.

A (85:58)

Seems doable. And then never bet against Elon, bro. Never get that. Never. Never been against. Come on. Anyway, Fin AI, the number one AI agent professor service number one in performance benchmarks. Number one competitor Bake Offs number one ranking on G2. There are so much more in the top timeline. We are going to have guests joining us in 10 minutes, but let's run through some more stories. Oh, the other story. OpenAI projected its cash burn this year through 2029 will rise even higher than previously thought to a total of 115 billion. That's about 80 billion higher than the company previously expected. Andrew Cote says, now that's what I call a nonprofit. This, the numbers are big, but. But I'm so inured to big numbers at this point, I'm like, yeah, that seems fine. I don't know. That's my take is that that seems completely appropriate for the opportunity you're building. The next big, massive consumer company. Everyone loves it. You have strong product market fit. All your competitors are kind of bowing out, more or less.

B (87:06)

The hyperscalers are bowing out.

A (87:08)

I would say that they're bowing out. A consumer.

B (87:10)

Yeah, I don't know. Nano Banana's ripping.

A (87:13)

Nano Banana is ripping. Yeah. AJ Ajay Midha at a 16Z. Is that his name? He had a good post about this today, talking about the revenge of the empire or something like that. Was that in here? Where was it? Ajni. I have him in here somewhere. I gotta get a better tab management situation going. Okay, here we go. Ajni Miha at Andreessen Horowitz says the empire strikes back and shows ChatGPT versus Gemini ChatGPT interest over time. On. Oh, well, I mean, you're gonna trust Google on how Gemini is doing in search.

B (87:55)

Also, there was other reporting that's. There was reporting that showed that agents are completely throwing like agents that are crawling the web are totally throwing trends.

A (88:05)

Interesting. No, I was completely kidding about Google cooking the results here. Obviously they wouldn't do that. But Nano Banana launches and Gemini is definitely mooning again. I see them as two separate use cases. When I think about the Nano Banana product, I don't see that as a direct competitor to GPT5 and the router to just answer questions, do things for you. Nanobanao is incredible. Gemini clearly has a huge advantage in video and image generation. That's why we're excited to partner with Fall because they are a vendor for that and we'll help you get set up with it.

B (88:50)

Most businesses don't necessarily care about. They're not looking to pick a single model. If you're a company like Canva, which uses Fall, they're leveraging in multiple different models at any given point.

A (89:06)

So if I think about, like, what does nanobanana mean, I feel like it is an incredible product that will be baked into YouTube. You'll be able to generate thumbnails within the YouTube studio. There will be products that people use to generate imagery and that will go out everywhere. If you're in an email and you need to generate imagery, like it'll be baked into Gemini, into, into all the products. But, but this particular spike in the chart reads to me as, as, as if a studio Ghibli moment where it's like you get a ton of attention and people come in. But like, what is driving ChatGPT use right now? It's not Ghibli's. That's not why people are coming to it. Like, the image generation is, is definitely used, but I would say it's probably less than like 10% of queries. Probably less than 2% of queries. Honestly, just based on my personal use, most of the time I'm looking for a fact, I'm looking for a breakdown, I'm looking for some analysis, go search the web, put something together and then every once in a while I go to it and I'm like, okay, now I need to do an image generation thing. And so I don't necessarily fully agree with this. Take that the empire is striking back. I mean, it is a strike. They're striking back. But has the empire won? We don't know.

B (90:15)

Good question.

A (90:16)

Anyway, question for you. Isaac says, I don't know anything about watches, but I'm looking for something respectable to wear on my wrist. Something that a watch nerd might go nice at but won't break the bank. Any recommendations? What you got, Jordy?

B (90:33)

What is not breaking the bank mean?

A (90:35)

Richard Mille.

B (90:37)

Yeah.

A (90:37)

Stay away from the FP Journes and the piece uniques because that's going to break the bank. But nice RM a racing machine on.

B (90:44)

The sports watch like a. Like an F. F.P. journe.

A (90:48)

That could work too. Yeah, you're going to want to stay away from the Graff diamonds hallucination because that's up in the 55 million range. But a Paul Newman Daytona, that's going to be something the watch nerds going to go.

B (91:00)

Fortunately, this is good news in the watch world. Trump was invited to the Rolex suite at the US Open men's final and he got to hang out with the Rolex CEO Jean Frederic Dufour. And the timeline is expecting some potential resolution of relief from the Swiss tariffs to come this week. Obviously tariffs impacting at least secondary watch prices right away. They've been Kind of up across the board. So good if you have a bunch of watches you want to sell, bad if you're in the market.

A (91:33)

Well, if you're looking for a watch, go to getbezel.com, your bezel concierge is available now to source you any watch on the planet. Seriously, any watch. You can chat with the Bezel concierge and you can ask them that exact prompt. You can say, what's a watch that would make. Make a watch nerd, say nice but not break the bank. They might recommend, I don't know, GMT Master, Batman, Rolex, something like that. That might be more reasonable. Anyway, Steve Jobs office versus Tim Cook's office. I think this says a lot about why Apple feels different under cook, says Sherman McCoy. But. But Sherman got put in the truth zone because that famous picture of Steve Jobs office where it's all messy and Tim Cook's office is all nice and clean. Well, guess what?

B (92:17)

See Tim Cook's home office.

A (92:18)

Yeah, let's see Tim Cook's home office. It could be 10 times messier. Imagine he's just stacked up.

B (92:23)

He's just the biggest horror.

A (92:24)

Yeah, ten times the books, just stacks of papers everywhere. So the picture on the left, it is Steve Jobs, but that's his home office. And Tim Cook is of course at. At the work office. But it's still probably something true there. Obviously Tim Cook is the operations mastermind. Tim Steve Jobs is more the creative genius.

B (92:44)

And so we got an old email here. February 13, 2005, Sergey Brin sent an email to his executive management group. Read the subject line, subject irate call from Steve Jobs. And then he just types this out clearly, just like stream of thought stream, stream of consciousness. So I got a call from Steve Jobs today who is very agitated. It was about us recruiting from the Safari team. He was sure we were building a browser and we're trying to get the Safari team. He made various veiled threats too, though I'm not inclined to hold them against him too much as he seemed beside himself, as Eric would say. So I just wanted to check what our status was in various respects and what we want to do about partners, friendly companies and recruiting on the browser. I know and told him that we have Mozilla people working here, largely on Firefox. I did not mention we may release an enhanced version, but I am not sure we are going to yet. On recruiting. I've heard recently of one candidate out of Apple that had browser expertise, so I guess he would be on Safari. I mentioned this to Steve. And he told me he was cool with us hiring anyone who came to us but was angry about systematic solicitation. I don't know if there's some systematic Safari recruiting effort that we have. Anyhow, I told him we are not building a browser and that to my knowledge, we were not systematically going after the Safari team in particular and that we should talk about various opportunities. I also said I would follow up and check on our recruiting strategies with regard to Apple and Safari. He seems soothed. So please update me on what you know here and on what you think we should have as a policy. On another note, it seems silly to have both Firefox and Safari. Perhaps there is some unification strategy that we can get to these two to pursue. Combined, they certainly have enough market share to drive webmasters and they would in fact build their house.

A (94:34)

The phrase in here, systematic solicitation, it's like that defines the modern tech era so perfectly. There was a time when CEOs would call each other and be like, hey, if. If my people come to you, that's cool, but just don't create a list of everyone who works for my company and then try and go poach all of them simultaneously in one weekend while we're on vacation.

B (94:57)

Don't call them personally, text them, don't.

A (95:01)

Invite them to dinner, invite them on.

B (95:03)

Your yacht, offer them $100 million.

A (95:05)

Don't go wave surfing with them in.

B (95:06)

Lake Tahoe if they don't accept 100, offer 200, offer 200. Just keep going.

A (95:11)

Please don't do that.

B (95:12)

But it is interesting. Three and a half years later, they launched Google Chrome.

A (95:15)

Wow.

B (95:16)

So it is interesting. In this email, you would think now, given what we know about platforms and operating systems and data, you would think that even at this point, Google was thinking we want to own the browser. But if, obviously. I'm sure when he wrote this, he wasn't expecting it to one day be public.

A (95:37)

Yeah, we got to dig more into the browser wars. We were talking this weekend about perplexity.

B (95:42)

Are they over?

A (95:43)

I don't know. I think there's going to be another run at it for sure and we're going to see some developments. Anyway, if you're trying to poach engineers from big tech, you know the best way to do it. Billboard on the 101. Go to adquick.com out of home advertising made 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 efficiency in this advertising.

B (96:06)

And just find the billboards and just surround your enemies with recruiting messaging.

A (96:11)

Yeah. Or pull a Schiffman and buy 300.

B (96:14)

Bill said the largest.

A (96:16)

That cannot be.

B (96:17)

Trust that.

A (96:18)

That cannot be true.

B (96:19)

It could be true on a pure. Seems impossible. But on a pure volume basis. Like just like volume of every. In a single buy.

A (96:29)

Yeah. Maybe for like a single creative or something. Like a single image is the biggest. But like, like when Apple launched the new iPhone, they put billboards in every single city and it's huge. And I see them everywhere.

B (96:40)

It definitely is.

A (96:40)

And movies, movies come out. There's billboards everywhere.

B (96:44)

It's the most YOLO marketing move that potentially of the year.

A (96:48)

It caught my eye and it made me think, I gotta buy one. We gotta demo it. Whoa. Make Tyler live with it for a week and see what happens to him in the guinea pig.

B (96:58)

The question is, could you use that money a lot more efficiently? You know, buying ads on meta, you know, it's a consumer product. Probably easy to figure out like who's the target buyer.

A (97:09)

Yeah.

B (97:09)

You can get the halo effect of out of home advertising by like buying strategic inventory.

A (97:15)

I do, I do think that there is something special about the having friend.com. so like when you go to friend.com, you just see the pendant and that's it. And then you scroll down and you see this like Apple, like experience. It says friend and then you scroll down and it has these questions. And so if I, if I put, if it was, if it was tryfriend AI and you put that on a billboard, I don't think that converts as nearly as well as friend.com.

B (97:45)

The domain is the one thing that I think in the fullness because I think the company doesn't work in two years. We can just wind the company down. He, he, he. And you have to look at the domain as like, like, does it increase your probability of success? I think for something like this, like an always on listening device, it's already. People aren't going to trust it. So domains can deliver trust and then does it.

A (98:09)

Yeah.

B (98:09)

Does it just increase conversion rate of people just caring enough to, to not, not to mention just like seeing it's a great domain for out of home advertising, period.

A (98:19)

Yeah, no, no. I truly think it unlocks out of home advertising in a completely different way that tryfriend AI would just convert way less than friend dot com. I want to know what's on that website. I'm just going to type in friend.com and then I'm on the flow. So I'm Actually surprisingly bullish on a big, big campaign. A big out of home campaign for friend.com. good luck to them. Well, we have our first guest of the show. We have Sunny from Grok. Infrastructure for inference, purpose built for speed, quality, constant.

B (98:51)

Welcome to the show.

A (98:51)

Thank you for joining the show today. How are you doing? Welcome to the stream. I'm doing good, guys.

G (98:56)

I'm actually at the all in Summit, so it's like, you know, coming in.

A (98:59)

Yeah, exactly.

G (99:00)

So I'm just in a green room here. So excited to be on with you guys.

B (99:04)

Did you already talk or is that coming up?

G (99:06)

No, no, no, I'm not talking.

E (99:07)

I'm just trying to like.

A (99:08)

Oh, but still in the green, still.

B (99:09)

In the green room.

A (99:10)

Okay.

G (99:12)

I had to get a favor so I could come on with you guys here.

A (99:15)

So, so nice. I appreciate it. Yeah, take us through the last last week. I, I believe you were at the dinner. What was kind of the what, what was the, what was the vibe like? Just, just walk us through your experience.

G (99:28)

Yeah, I mean, really special. Like, look, so first things first, I would say, you know, when you think about, you know, the President, what he's really done is he understands that he needs to enable technologists to basically do their thing. And I would, I would summarize it in the following ways. One, he brought everyone in and said, look, if you have some kind of issue, there's a team of people that you can work with, including David Sacks or just himself, to get things unblocked. And there's many issues that companies deal with, whether it comes to export control or tariffs that are being put on or fines that are being put on our companies. And so he's really making himself available to help our companies broadly. I think he's also understood that, you know, AI and technology is the forefront of where many of the innovations for the country will happen, including huge infrastructure projects.

A (100:20)

Right.

G (100:20)

And so you heard that when the press came in and when they're asking the questions how much each company is spending, you know, looking at, you know, trillions of dollars of spend, which has downstream impact in manufacturing, construction and other places. And so, you know, that's something he's excited about. And I think lastly, like this, we really want America to win. And he wants America to win. And so he's basically saying, look, I've got your back, I'm supporting you here. And I'm basically putting this on the forefront. So that's the high level. I would say the experience, just diving into it was really special. Started with a small gathering on the Rose Garden Club outside, it was raining. We were supposed to have a dinner there, but it got moved. And then a bunch of us got, you know, all of us, I got moved into the Roosevelt Room, which is connected into the, into the Oval Office. And then, you know, he took his time basically meeting with each one of us quickly getting everybody gathered together and just kind of understanding concerns that folks had. And, you know, basically us getting an experience in Oval Office. We all got a challenge coin, we got a pen, which is really cool, like a little, you know, a memento, a very special memento. And then we all went to the East Wing for a dinner and private conversation followed by a little bit of the press came in, which you guys would have seen the videos and then conversation afterwards. And he really took the time to address everybody that was at the table and make sure that if anybody had a concern, they had a chance to surface it there.

B (101:49)

What was top of mind for you specifically?

G (101:53)

I think, look, the American AI stack, that's really important. That's something that was published at the AI Action Summit, a couple based, basically middle of July. And so what we're really focused on is basically making sure that we help the government understand what that stack looks like. And so I don't know if your producers can pull it up, but we published something a few weeks ago, maybe two weeks ago. Now that basically is our version of the stack and we think it's been positively received by the government and the administration. In terms of framework, we're not saying this is everything, we're not saying it's comprehensive, we're not saying it can't change. But really, you know, the, you know, kind of the Department of Commerce today is out there trying to understand what it should look like. And this is one view of the stack. And really what does that mean? Let's kind of double click into that. The stack is important because if the US is going to keep export controls so that we can control, you know, highly sensitive technology getting into the wrong hands, you need to look at it from a stack perspective. A chip alone is not something enough that you should basically, you know, put a control on. So, so you got to look at all the pieces. And on the flip side, if you are going to be a partner of the US and you're going to want to buy US assets, you want to make sure that you have the whole list of things, because if you don't have all the pieces, you're going to end up with paperweights, really heavy and expensive ones. So you need to have pieces along all. And that's what we sort of took a stab at when we put that out there.

A (103:21)

Speaking of the AI stack, can you help me understand at a more precise level how you're positioning the company or at least telling the story around the company in terms of where you fit in the stack? Because we saw last week OpenAI's doing a deal for custom silicon with Broadcom and then semianalysis came out with this article saying that basically Anthropic and AWS and Trainium are like super tightly co designed now Google obviously with the TPU and then I was, I was chatting with some OpenAI people, they're like, you know, we actually talk to Nvidia pretty closely. Like we're like, we like, I don't know if code design's the right word but like we give input and you know, the next generation of Nvidia GPUs will be very capable of running ChatGPT. And so, so, so where, how do you position where you fit into the, the American AI stack?

G (104:19)

Do you guys mind if I pull up a quick squeeze screen share?

A (104:22)

Yeah, you can, but we are live so anything you share will be shared. So do not share your private key to those crypto holdings.

B (104:29)

There you go.

A (104:30)

Thank you.

B (104:30)

Smooth.

G (104:32)

Yeah, yeah. So the way to think about your.

B (104:35)

Question exactly, could you zoom in a little bit? Is that possible? Or maybe we can on our side.

G (104:39)

Yeah, maybe you have to do it on your side. I don't know if I can, but I can try. You know really where that's called out at the bottom is. So just to answer your question first and I'll try to zoom in with.

A (104:52)

A. I think we can pull it up on our side. Yeah.

G (104:56)

So really at the bottom there, which is, you know, this is infrastructure and the bottom is compute. So in the compute bucket right at the bottom here you see Nvidia, Groq, amd, Qualcomm, Broadcom. Like I said, this is not comprehensive. You know, Broadcom is helping companies like Google and OpenAI and others make their chips.

A (105:12)

Right.

G (105:12)

And so the idea, you know, to answer your question, is that the models come higher up. It you should be able to look at the stack and say if you picked a piece from each one of these different categories that we've highlighted, you can have an end to end AI solution to either train a model to either inference a model or to build an application on top of it. Right. So those are the three things that this stack should enable. And you may not need all the pieces depending on what you're doing.

A (105:38)

Right.

G (105:38)

If you're trying to train a model, clearly you don't need databases. Right. But these are all the pieces that you need to basically be able to do something in AI.

A (105:46)

Got it. Fascinating. Yeah. Thank you for putting this together. This document is massive and deserves its own deep dive. We'll have to dig in deeper.

G (106:00)

We can do another one if you guys ever want to.

A (106:02)

Yeah, of course.

B (106:03)

What's the update on the projects in Europe? And then what do you guys have cooking stateside?

G (106:11)

Yeah, we've launched close to 14 different data centers this year, including one that we announced publicly. Mad decent there. Yeah, we've launched 14 this year and we're continuing to expand our footprint. Europe is really fascinating. You guys saw the announcement with ASML and their investment into Mistral, so it looks like they're waking up and starting to basically put resources behind these things. So I think that's great. Our focus primarily in that region of the world is centered around data center and arrangement that we have in Saudi Arabia. And so that's in partnership with Humane Aramco Digital, where it's the largest inference cluster in the Middle east and we believe all the way into Europe. And so we serve sort of most of Asia, Europe and the Middle east from that data center. And our goal right now what we're cooking up is just an expansion of that.

D (107:11)

Right.

G (107:12)

So we're working to expand that cluster, add more capabilities and basically more capacity to that cluster because it's been well received and well consumed.

A (107:20)

Very cool. I know you have a heart out, so we'll let you go. But last question for me on that Mistral ASML deal, it kind of jumped out to me because it feels like they're almost like jumping a layer of the stack. I typically think about, like ASML sells to tsmc, which sells to Nvidia and some other folks, and then the application layer and the foundation model labs sit on top of that. So you're kind of a couple steps away. Do you have any insight into what they were thinking? A lot of people were just saying, hey, it's a national champion. They want to support and they're willing to kind of go outside the typical business that they do to make an investment and. Or maybe they just see it in purely financial terms, but it seems like if it was purely financial, they would have. There's a bunch of other things that they could invest in, you know. Yeah, they're not a hedge fund. Yeah, right.

G (108:09)

Yeah, yeah, no, totally. I think the points you talked about are Valid. I would add one thing to it, which is if you actually go back and listen to the launch of xai, what Elon talks about is like, you know, companies and his companies primarily using, you know, AI to better themselves. Right. And I think if you're someone like asml, if you want a close partner, doing that with Mistral is quite smart, right? How can they improve the tools that they make so that we can get better chips, better stuff in that layer of the stack? And so I would say that's something that is getting underappreciated today. But more and more companies are doing that, right? Getting more advanced AI so that they can basically do that. And from their perspective, they'll never be able to do that their own. To your point, they're too far down. So the partnership there, maybe we'll get better machines, maybe we'll get better power consumption from those things. Maybe we'll get, you know, better lithography from them. So that would be something that, that I, I would, you know, if I was building a business outside of, you know, GROK today, that's something I would focus on is how can I really not just use this for the high level use cases we've seen, but really have it, you know, find new materials for me, look at, you know, different aspects of physics. And those are really hard problems that you need to be very closely embedded with a model maker, like a foundational model maker, if you really want to pull those off.

A (109:26)

Fascinating. Well, we will continue to monitor the situation as always and thank you so much for hopping on. Enjoy the rest of the insights on.

B (109:33)

Dinner and the update.

A (109:34)

We'll talk to you soon. Have a great rest of your day.

G (109:36)

Thanks for having me, guys.

A (109:37)

See ya.

H (109:37)

Cheers.

A (109:39)

Let me tell you about customer relationship Magic. ADEO is the AI native CRM that builds scales and grows your company to the next level. And get started for free with adeo. Theo says.

B (109:52)

One thing I wanted to highlight quickly because we didn't get to it earlier. T Mobile, AT&T and Verizon all dropped pretty meaningfully on the EchoStar news.

A (110:02)

Yeah, that makes sense because Elon's basically going around T Mobile. It won't need to go through them because they'll own their own piece of the spectrum.

B (110:11)

T mobile's market cap.

A (110:14)

20 billion.

B (110:15)

Close.

A (110:16)

How much?

B (110:17)

273 billion.

A (110:18)

273 billion. Wow. Beast. Okay.

B (110:24)

Bigger than 18. So Verizon. Yeah, good question. No, Verizon is 180, 82 billion.

A (110:30)

I didn't realize T mobile was bigger.

B (110:32)

Than at T is 206 billion. And Verizon is. Is bigger than all of them.

A (110:37)

Wow. It's almost like owning a monopoly is valuable.

B (110:41)

Yeah.

A (110:42)

Seriously, once you get those licenses, it just prints.

B (110:45)

Yeah. It's so crazy that EchoStar was like, on the verge of bankruptcy, like, not making debt payments, and they just like yolo. And they weren't even leveraging the spectrum that they had.

A (110:55)

No, they just owned it. Fascinating.

B (110:59)

You got a post from Fiat.

A (111:02)

I'd rather talk about this Andrew McCallop story. So project Bob has launched. You can go to Dashboard Dot, projectbob xyz. Christian Kyle says this is easily the coolest thing happening in the world today. Hobbyist launching an autonomous drone ship, hoping to circumnavigate the globe. Andrew, of course, works at Varda Industries on manufacturing in space, but in his free time, he tried to replicate the LK99 superconductor. I was there on the night that he tried to get the rocks to float. It was very, very interesting. He's just a fantastic scientist and builder. And I guess in his free time, he went and built a boat that drives itself, talks to Starlink and a few other networks and wired it all up. And who knows, maybe it turns into a business. That's the nature of launching side projects. If you're really good at them, eventually people say like, well, I'd like to buy some, sir, and Maybe, maybe the DoD will come to him. Or the Dow, as they're called.

B (112:00)

I'm so, I'm so excited for the drone era of exploration. Right. Just being able to send something off and, and follow it along live virtually is so cool. I always, like, I want Andrew to take this and properly send Antarctica, apparently, and just see, like, how thrilling would it be to have nothing for days and weeks and months on end and then. And then somebody just comes up on the video and just takes it out. What are they doing down there?

A (112:29)

Anyway? We have our next guest, Scott Wu from Cognition, joining the stream. Welcome to the stream. How are you doing, Scott?

F (112:37)

How's it going?

A (112:37)

It's good to see you again.

B (112:39)

It's been a little bit.

A (112:39)

Great to see you. I think I recognize that sweatshirt. Is that a founder?

H (112:43)

It is the founder.

A (112:44)

It is the founder.

H (112:45)

It sound like the right thing to do today.

A (112:47)

We interviewed Alex Karp last week and he referred to it as the Founders Fund, because back in the day it was actually called the Founders Fund. And then they dropped the the. Which I think is funny. Anyway, we're not talking about Founders Fund. We're talking about your Company. Give us the news. What's the latest?

H (113:05)

Yeah, so we just closed a big fundraise led by Founders Fund, actually. So we kind of are talking about founders funding. We closed a raise at a $10.2 billion valuation. Super grateful for all the support that we've had.

A (113:22)

Congratulations.

B (113:25)

Yeah.

H (113:26)

Busy month and a half for us since the Windsurf news. I mean, it's been a crazy couple months for code and for us especially, getting to just see the two products mesh together has been a lot of fun.

A (113:38)

Yeah. Where is the growth coming from? I remember seeing a presentation at Microsoft Build and they were talking about the value that Devin can bring to replatforming and that was super concrete for me because I can just imagine having an enterprise application written in. Net and wanting to move to Python and that being a lot of work to write the test suite and move everything over and just being able to unleash a ton of AI agents and coding agents seem like incredible value and an incredibly high leverage use of technology. Is that a big piece of the business? What else is changing? What else is growing in the last year?

H (114:19)

Yeah, I think that's right. A big part of what we do is I kind of describe it as going and taking on engineering toil, you know, and so that includes like all these RE platforms and migrations that folks have to do. That includes version upgrades, testing, documentation, you know, issue triaging, all of these various things. And it's, you know, the reality is that that's a lot of what takes up time for software engineers today. I don't think it's anyone's favorite thing to do, but it is, it is a lot of the time for people.

A (114:45)

Right.

H (114:46)

And so that's often what we see. And you know, we've been working with a pretty big range of companies all the way from the smallest startups all the way to some of the biggest enterprises in the world.

A (114:54)

But yeah, yeah, because Devon is in ga, so any startup can sign up for it. Correct. But then obviously you have a very talented team that goes and sells into very large enterprises as well.

B (115:07)

Yeah. What's it been? Yeah, Windsurf was known for their GTM team. What's it been like? Just like bolting on or really integrating a team like that, you guys. Yeah, I'm sure it's really added a lot of firepower.

H (115:22)

It's been incredible, honestly. I mean even I think just the two products themselves because, you know, as we were saying last time, like a lot of the. There's kind of these two categories of product experiences that devs are using today. Right. There's the ides and the agents, and even just being able to have both and kind of serve that all in one solution with Windsurf and Devon has been amazing for us. And so we've seen a bunch of. I mean, Windsurf and Devin were both already each individually growing, but in the last month and a half, half, it's grown even faster.

A (115:50)

I want to go into a little bit of a debunkathon, hit you with some data points and you can give me a lot more context. So the first one that we saw, this is all from extremely reliable sources, usually random screenshots that I see on X. But one was that when a developer uses an AI coding tool, they generate twice as much code, but 10 times as many security vulnerabilities. Do you have anything to add? Does that sound right? Does that sound wildly off? What can we do to stop creating security vulnerabilities?

H (116:26)

Yeah, there's a lot of studies out there. Studies have different sample sizes and they have different ways that they measure these things for us with a lot of the projects that we deliver. Because if somebody's doing a whole re platform or something, that is a concrete project that you're. You're setting up a team to do and kind of accounting for how many hours you need to do. With the enterprise customers that we work with, we typically see speed ups of around 8 to 15x in terms of how long it takes. Yeah. So basically they're doing in one hour of an engineer's time using Devon what would typically take eight hours without Devon. And I think the main thing that you really have to call out here is it really depends on the use case. And so I think for a lot of these kind of like repetitive, tedious things that frankly the engineers don't want to do anyway, that's where tools like Devin work really, really, really well. Devin is not necessarily. You could use Devin to try and kind of like be the ideator with you or something like that, but it's obviously it's not today what we've optimized Devin for. And so that's perhaps what folks are seeing in some of these different coding products.

A (117:29)

Yeah. And then on the security vulnerability side, do you need a, you know, okay, you have Devin, but do you need like Bill, like a different agent that just is watching for security only really in traditional software development you might have like, yeah, there's this great developer that can just churn out code really fast. But then we got the guy over here or the gal over here that really knows what is secure and what's not and where the flags are. And you have pen testers like is that a different product? Is that just a different layer of what you're training for? Or have you, have you kind of built Devon in a way that you don't inject security vulnerabilities?

H (118:08)

Yeah, it's a good question. So I think with security, I think the main thing I would just call out here is like this is a problem that all engineering orgs have already had to think about and deal with. You know, it's not necessarily new in the age of AI that you have to think about developers introducing security vulnerabilities. Right. And yeah, you know, to your point, we have a lot of, of these processes in place for that. Right. Which is, you know, the first one probably is just code review itself. Right. I mean people, you know, typically you, you don't allow people to just merge their code straight to master or, or, or, or something like that. And, and similarly, you know, you want to have similar processes in place for your AI agents. I think there will be specific kind of like security based agents, but I, I kind of imagine it draws on a lot of the same code based intelligence and just like understanding, you know, what these functions are meant for or how these are meant to be used, which is kind of a shared intelligence. And so maybe we release our own kind of like security theme to Devin that specifically does this security review or something like that.

B (119:05)

Are you seeing companies change their hiring practices at all after implementing Devin and starting to get value out of it? We've just been curious to learn how to. Companies are actually adapting as they sort of ramp up AI coding tools.

H (119:24)

Yeah, it's a great question. It depends a lot on the space of company because obviously all companies are competing with the others in their space and that's what ultimately comes down to this. But honestly, I think what we've seen actually it looks like folks taking more and more projects in house. And so a lot of things that folks typically would have outsourced to a team of contractors or, or handed off to a system integrator or something like that, they're actually just doing in house with Devon and they're able to do that faster. One of our biggest customers actually is a bank in Latin America that started using us around the start of this year for some of these refactors and migrations and so on, and then just got to the point where their team was seeing a lot more effectiveness using the tool and then they 10x their contract within the last 8 or 9 months. And if anything, they're growing headcount. They're not shrinking every single one of their. If every engineer is worthwhile more, then obviously you want more engineers.

B (120:19)

Add more engineers.

A (120:20)

Yes. Okay, let's continue the debunkathon. We saw some data from the census that showed that AI is cooked. It's done, it's over, it dropped. The actual data point was for firms polled by the census that they have over 250 employees. The level of AI adoption had dropped from 414 to 12%. A little dip in the chart. Of course people are joking about this, memeing about this, you know, saying it's all over, but also joking that of course this is just one data point. To me, it seemed extremely low to think that only 10% of businesses are even using AI, since AI is kind of everywhere now and everyone uses AI and all sorts of things. Anyway. What is your take on, like, what's going on in that data point or in that stat specifically? And what other kind of context can you add for us?

H (121:16)

Yeah, I think I saw this chart. This is the one. This was going around Twitter where it says like, in July and August there was like a tiny, you know, and then everyone was freaking about.

A (121:23)

Yes.

B (121:23)

Yeah.

H (121:24)

So saying, you know, I don't really, I haven't looked into the methodology of how they figured out that number. It's kind of interesting because, I mean, for us, obviously, like a lot of our work is with big enterprises, you know, specifically working on these massive code bases. And July and August have been our biggest months ever, you know, by far. So that's, that's, that's, that's in terms of the data. But, but I feel like the high level thing which I would just call out is, you know, I think there's still a kind of like, like collective, like, I don't know, there's like a, there's like a collective, like refusal to see, you know, reality. Which is, if I were to say it is Basically, we have AI that passes the Turing test. We have AI that gets an IMO gold medal. We have AI that does all these crazy things, which 10 years ago we would have put it clearly called AGI. And it doesn't mean that you do all of this work for free as a result. There's obviously still a ton that you have to do to teach the specific capabilities and to make the right product experiences and to get actual businesses out there in the world going and using these things. But I just think the story know, the things that we're asking the AI to do are not harder than getting an IMO gold medal, put it that way. And so there's, there's a lot of work to do. You know, I don't want to diminish. Like it's going to be years and years, you know, of taking all of the techniques that we now have and figuring out how to build the right experiences and get them out to everyone. But I really don't think. Also, I love that every time I come, I'm just the, I'm just the crazy AI bullet, you know, I knew. But I think it seems very clear that AI will do more and more of this work over time. In software engineering, I think you see this especially, which is, if anything, I think the best coders are the ones whose flows have changed the most dramatically. And you just think about how many engineers there are out there that still have to learn these tools and have to get on these things. I don't think we are due for a slowdown anytime too soon.

A (123:20)

Yeah, and the methodology of the census data was a little bit odd to me. Basically they called up individual proprietors, business owners, and they said, are you using AI? And they defined AI as predictive analytics, data analytics, image processing, voice recognition, machine learning, natural language.

B (123:42)

So a bunch of people said, no, we're not doing data, we're not doing data analysis.

A (123:46)

Which was very odd. Only 10% of people said yes to this question. I feel like it should be like 100% because like if you, if you're using Stripe, like there's machine learning under the hood for fraud detection. If you're using Gmail, like Gemini is right there, there's going to be autocomplete on your phone is using data analysis, like, anyway, my real question is one of the, one of the lowest hanging fruits that I see in the software that I interact with is in government, you know, government websites, basically. I imagine that the census could benefit from having Devin run around, write better polling software. Is there anything on the horizon where you might be working with the government? Is that something where you need to get ITAR compliance or something before you go in there? Is it on the roadmap? What do you think in terms of improving the software that our government serves to us?

H (124:34)

Yeah, no, I think it's a great point. And as you say, I think there's a lot of work to do and it's a great example of the kind of like the messy problems with which I think it is very clear will be solved. Over the next couple years. There's no reason that the government shouldn't be using tools like these to go and build their websites better and faster. And everyone knows how many bugs there are, how many simple things that could just be fixed if there were an AI software engineer available to go and do this. It's funny because I think Windsurf actually, funnily enough, was ahead of us on getting things moving with Fedramp and getting certified in all these things. But of course. But it's something that we're working on as well.

A (125:11)

That's very exciting.

B (125:13)

Are you thinking at all about other acquisitions over the next couple years? You got a big balance sheet at this point? Yeah, yeah.

H (125:22)

I mean the last one went pretty well. I think for us it's obviously really valuable to have this capital for a few different reasons. You know, with model training, I think with hiring great talent, with expanding go to market and so on, you know, is it possible that part of that will also, you know, go towards, you know, buying other companies? Of course it's possible. There are not any plans at the moment, but yeah, yeah.

A (125:49)

In terms of your position in the stack, I see you as sort of like a initially like a enterprise level application layer company that then is now going down with Windsurf into more of a B2B context. Not full consumer, but more broad usage. But famously model agnostic, not training foundation models not working down the stack. And we're seeing the foundation model companies do deals with Broadcom to co develop chips and what Elon's doing with Samsung. And there's a lot of these data points. Are there any other places where you want to at least have better relationships with other layers of the stack? Is any of that important or is it really just like you have your slice and you're just going to go and run at that for the foreseeable future?

H (126:41)

Yeah, it's a really good question. I mean, I think in terms of relationships for sure. I mean we work very closely with the foundation labs, for example. We have lots of things going on in terms of how we do compute or even kind of like partners that we work with on distribution, which I think are great. I would say, you know, I think the. For us it's. We, we obviously, I think it's very important for us to just have a lot of humility about where we're coming from and what we're doing. You know, we're obviously, you know, a very new entrant to the space and you know, are only like 0.1% of the way there. And I think, you know, for us to be able to meaningfully succeed is going to require a lot of focus. And so I think from that perspective of, I think the like, we know what is the area that we really kind of want to own. And that's basically everything between that goes from taking the model, the base model itself, and then doing the right kind of capabilities work and building that into a product experience that real engineers can use every day. And I think we will stay quite narrow kind of within that focus ourselves for this.

B (127:43)

So you're saying job's not finished?

H (127:49)

I mean, it's barely even started, honestly, I think. And I think obviously there's, I think, a vision of the future where, as we said, where software engineering is just as easy as telling your computer what to do. And I think the world of software engineering has already changed in some meaningful ways over the last two, three years because of all these AI coding tools. But I think, I think we've got another, like, 10, 20 generations of product experiences as a community to build and unlock until we get to that full future.

A (128:22)

Important question was this latest round of financing negotiated with Founders Fund over poker or chess?

H (128:34)

Unfortunately, it was neither. It was funny, actually. We were on the weekend that we did the windsurfing, obviously. I mean, it was a pretty substantial commitment from us in terms of windserve. And so we explicitly needed board approval, but also wanted to give our investors an update and kind of hear what they thought and so on. And so we were on with the Founders Fund team with Napoleon and Peter and getting their thoughts on what they thought. Makes sense. And I mean, first of all, they were really supportive of us doing the deal, and they thought the partnership made a lot of sense for a lot of the same reasons that we did. But they also were very clear of, like, hey, I know this is going to be a big thing for you guys on the balance sheet and.

A (129:15)

For.

H (129:15)

You guys as a company, and we want to be clear that we stand ready to support you in the next thing if this is the deal that you want to do. And honestly, that's a lot of what gave us the confidence to be able to go ahead with it and close the whole deal in the 48 hours or 72 hours or whatever it ended up being.

A (129:31)

And so.

B (129:32)

So we're good for our 400 million?

H (129:34)

Yeah, it's really been amazing. I can't shout them out enough. Napoleon and Peter and the whole team that we've gotten to work with have been great to us.

A (129:45)

That's fantastic. Tell us the story of the deal that was negotiated over. Was it poker? How did that come together?

H (129:52)

Yeah, so it's a funny one in retrospect. So it didn't actually happen, but basically when we were doing this, the Series A, you know, one of the very early rounds of the company, it was me and Napoleon and, you know, I was here and he was there or something like that. And there was like, I think at some point it was clear that the deal was going to get done, you know, but it was just kind of like going through some of the more minor terms and making sure we were on the same page of everything. And we kind of put, you know, what if we, what if, what if we just had a nice heads up match to go and settle this and decide whose terms we're going to take. We did not do that. Probably for the better that we didn't do that.

B (130:27)

But, but, but, yeah, yeah.

A (130:29)

Oh, yeah. But the story lives on as something it almost happened or was at least like joked about, which is fantastic. That is great. Well, thank you so much for taking the time. Jordy, do you have anything else?

B (130:39)

No. Congrats.

A (130:40)

I know you got a busy day, so we'll let you get back to it.

B (130:42)

Just try to appreciate, try to, try to have another reason to come on before the end of the year. At least a couple, you know, it's.

E (130:48)

A great lock in.

A (130:50)

There's going to be plenty more stuff to debunk. There's going to be tons of data points that people read into way too much and we will be giving you a call.

B (130:58)

I was going to say, I look.

H (130:59)

Forward to coming on again and being the idiot optimist that just, you know, everyone else is giving their rational arguments about this is going to work. This is not going to work. And I'm just here saying everything's bullish.

A (131:07)

But yeah, I mean, if you want one more, we can talk about inference costs. I mean, people are saying that, oh, inference costs aren't falling, you got to use the latest and greatest model and your gross margins are going to be terrible forever. Debunk that one for me. Sure.

H (131:22)

No, I mean, I think there's. So I think there's two separate questions there. One is the question of is the business going to be defensible? Are you going to be able to offer something that you're competing with your competitors on, something that is not just purely a race to the bottom? And then the second thing is, is the cost value trade off for the customers going to actually make sense? Right. I think the second one is just very obviously true because as these tools, if the trade off is doing more and more and speeding up labor, making every accountant faster and making every lawyer faster and so on, at some point it's pretty obvious that the machines are cheaper, cheap enough that making every lawyer three times faster is just obviously a no brainer. I think on the point of the first one, obviously it's a great point and I think that's it's different for every different company. I think a lot of the businesses in the application layer need to think a lot about what is their special sauce and what is their differentiation. But I think the answer for most probably of what that comes to is a combination of one is just really, really focusing on specific use cases and delivering solutions that are really kind of custom solutions for the problem that they're specifically solving. And then two is, I think there is a real kind of, I'll call it a personalization effect that I think we're seeing more and more. There's ChatGPT memory for example. Obviously Devin has its own whole knowledge system where as the tool just understands the customer and their own business and their trade offs and everything much, much better. That itself is obviously something that just makes the product more and more powerful just for them and is the kind of thing that would command that bad you. So I think that it's, it's, you know, from a perspective of is AI, you know, are we going to have monster intelligent AI that's just so smart. But we just, you know, we just, we just turn the machines off because they're too expensive. You know, I don't think we'll have that future. I think there's a, there's a very reasonable question about, you know, where that value capture comes in. And, and I think that's why all these businesses are thinking about, you know, what is this one thing that I'm going to do really, really well that's going to make my business terrible.

A (133:29)

It's fantastic. Well, thank you so much for taking the time to hop on on a busy day. Good to hear that the job's not finished. I love unfinished jobs.

C (133:37)

It's the best.

B (133:39)

I hate when the job's done.

A (133:41)

I hate when the job's done.

B (133:42)

Give our best to the team.

A (133:43)

We will talk to you soon, will you?

H (133:45)

Thanks for having me.

A (133:46)

See you guys. Bye. How'd you sleep last night, Jordy? I had, you had a terrible night, right?

B (133:52)

I had a brutal night but not because of my eighth sleep. Sleep.

A (133:54)

No, I got an 82. I'm sleeping pretty good. I think I get the sound effect.

B (133:58)

I managed to pull a 75 despite.

A (134:01)

Eat sleep is working overtime. It was like we gotta get this guy some sleep. There's kids everywhere.

B (134:08)

Three year old is up. The one year old throwing up we.

A (134:11)

Gotta it was just adaptive AI coming in to help you sleep better. Anyway go to 8sleep.com get a pod 55 year warranty, 30 night risk free trial free returns, free shipping and we will bring in our next guest ara Kharazian from ramp.com Time is money save both easy use corporate cars, bill pay accounting and a whole lot more. Good to see you Ara, how you doing?

B (134:35)

Great to see you.

C (134:36)

Thanks for having me again.

B (134:37)

Thanks for hopping on so much data. Data driven chaos in the timeline.

A (134:42)

Yeah, I want to talk about996 but first let's start with this data that it's completely over and no businesses are going to be using AI because if you look at the trend line we went from 12% to 10%. If you track that out on the line it's going to be zero percent. No, not that one. It might be negative. 50% of businesses using.

C (135:04)

You're talking about the AI adoption data from census.

A (135:06)

I am talking about the census AI.

B (135:08)

Adoption data folks that brought you the Department of Motor Vehicles. Yes, the real.

A (135:13)

What was your reading?

C (135:15)

I've been thinking about the census AI adoption data for a long time because we have our own measurement of it with ramp data. Yep, that's.com data aiindex. What I don't like about the census measurement and why I think it might be underestimating AI adoption is that the question that they have written is not the right way to measure AI adoption. Essentially the way the census measures it is. They ask ask businesses do you use AI to produce goods and services? And that makes sense because they wrote the question back in like 2023 when we weren't really sure what AI was going to look like. But AI to produce goods and services, that's first of all that's econ speak, right? That's how economists talk to each other.

B (135:54)

Well yeah. And even if somebody technically like in my view if somebody provides any type of services to somebody and they use AI call transcription like someone on their team is like recording calls and like transcribing with AI, that to me like qualifies as I literally had in 2013.

A (136:13)

I was running a consumer packaged goods company. Our business was making protein shakes. That was the physical good that we produced. But we also would transcribe all of our all of our calls. And so were we using AI to produce goods and services? I would have said absolutely. We're using, we're using linear regression and we're projecting out our financials and we're doing all sorts of stuff that, that fits in the bucket. But yes, to your point, you're a.

C (136:37)

Forward thinking business leader.

A (136:39)

Sure.

C (136:39)

And, and that makes sense. But most people reading that think, oh, am I literally using AI to produce widgets on my factory floor?

A (136:47)

Yep, yep.

C (136:48)

It's, it's the kind of question that doesn't really capture the way that AI in the past three or four years of its development has more or less has mostly been adopted by businesses for back office tasks.

A (136:58)

Yep.

C (137:00)

And when you look at that 10%, I think most people would say, yeah, for most businesses that's pretty low. Forget, you know, tech forward businesses just in general.

A (137:08)

Yeah. So the overall number of like 10% of American businesses using AI, that seemed ridiculously low to me because just consumer LLMs like ChatGPT are like 50% adoption amongst American adults. And so it just doesn't math with me that someone would be, that four out of five Americans would be using ChatGPT and then walk into work and be like, no, I'm not going to use AI for my job. But what I'm more interested in is like why do we think that there's a peak and then a fall off at all? My theory was like maybe these big companies are getting sold on. You got to use AI to produce goods and services. They try and roll out some zombie change management strategy where they're remember the.

B (137:52)

Karna, the Klarna CEO? Like, like did this thing, I'm going.

A (137:55)

To use AI for everything.

B (137:56)

And then he ended up saying take it back.

A (137:58)

Yeah.

B (137:58)

And so people are pretty, pretty good.

A (138:00)

Yeah. So did you have a read on that at all?

C (138:03)

Well, there's a few things. One, the sample sizes for these surveys are actually pretty small. They run these surveys every two weeks.

A (138:09)

Yeah.

C (138:09)

And you know, like most government survey collection, particularly if it's run every two weeks, you're going to get some mixed differences in who responds.

A (138:17)

Sure.

C (138:17)

So that's part of it. I think some of it just might be some noise in the data set because it did come back up in the most recent read. It wasn't that much of a decline in the first place. So you could write it off as, as a noisy data. The second thing is that it's not unreasonable to think that the fact that this was conducted in the summer might mean that even at the Businesses who responded to the survey. The person who responded to that survey at that business might be different.

A (138:42)

Yep, totally.

C (138:43)

This is another reason why I really don't like business surveys is that they try to capture what a business business is thinking, which is a large, complex organization, but they require one person at that business to answer the question.

B (138:53)

Easy for you to say. You've got tens of thousands of businesses and their actual purchasing behavior.

A (138:57)

Yeah. So flip it over. What is the data actually saying about the level of AI adoption in the business world?

C (139:07)

When we look at ramp spend data, AI adoption is up. AI adoption is now about 45% of businesses in the U.S. it varies by sector, so adoption is much higher in tech and finance. About 70% of tech firms on our platform adopted AI in some paid form. I still think that's probably pretty low. We're not capturing free usage or capturing employee usage on their own personal accounts. But then even in restaurants, for example, we see 20% adoption of AI, and I think that's likely.

B (139:33)

And what does that mean in the restaurant context? It's like that could be they're paying for chatgpts like Pro Plan or they're using a marketing marketing materials like developing.

C (139:44)

Like, you know, I mean, I've seen menus that have AI generated images. That's not my favorite example.

A (139:50)

Yeah.

C (139:51)

But a much better example is, is designing a Facebook ad that uses an AI generated image or copy that really speeds up a lot of the work that a restaurant owner otherwise has to do. I mean, these restaurants and firms often have really small staffs.

A (140:04)

Yeah. Yep.

C (140:05)

And so I've talked to some restaurant owners who use ramp, for example, and talk to them about how they use AI. And for them, for the most part, it does automate a lot of the sort of not just back office work, but a lot of the work that lets them get back to work of doing the restaurant work that they want to do.

A (140:21)

Okay, so let's flip it over to the996 phenomenon. There's been some spicy quotes on the timeline talking about how 9, 9, 6 working on Saturday is super trendy. Every everyone's doing it. What does the data say?

C (140:35)

So does everyone know what 996 is in the audience?

B (140:40)

I don't think our.

A (140:41)

I think everyone knows.

C (140:42)

But break it down for us.996 is associated with Chinese working culture, though I believe it's actually illegal in China now, though I think it's still often practiced. And it means the work 9am to 9pm Six days a week.

A (140:58)

Yeah.

C (141:00)

And then it Kind of seems like it's caught fire in Silicon Valley and tech culture in this whole like wellness oriented run fast, lift heavy, MARRY early work, 9am to 9pm Six days a week and just focus on your business. Most of the stories have been vibes based like it's just people in startup world talking about what they're doing. It's true, it's actually happening and shows up.

B (141:25)

The question is, how will you know a founder works 996?

A (141:28)

They'll tell you.

B (141:29)

They'll tell you.

A (141:31)

Yeah, but I mean a lot of people were wondering, is this a larp? Is this something where, yes, Saturday is for working, it's for posting about how you're working. But you're saying that the data actually suggests that there is business activity happening on Saturdays in startups at an increased rate?

C (141:50)

Well, in the one specific way we looked at it in this one, we looked at who's, you know, employee cards. So we're not necessarily just looking at founders, we're looking at just people at the company who are more or less expensing business meals or like late time meals on Saturdays or like doordash takeout, other things like that. A lot of the criticism that I've gotten today on the timeline has been like, oh well, you can't tell if maybe some people are saying it's fraud. It must be these people who are using their business cards. Like I want you to read my post because in my post I found it. It's just sf. Oh yeah, this isn't happening in New York.

A (142:27)

Interesting.

C (142:27)

This isn't happening in Austin.

A (142:29)

Interesting.

C (142:30)

It's not happening in Miami. This is just an SF thing. And it's also really new. This wasn't happening last year.

A (142:36)

Wow.

C (142:37)

Perfectly coincides with when people started talking about the 996 cultural movement.

A (142:42)

Fascinating.

C (142:43)

Unless you want to argue that, oh, fraud's this whole.

B (142:45)

I mean it really proves. It really proves there's no such thing as a free lunch. Because in these cases the employees are getting a free lunch but they had to work on Saturday.

C (142:54)

It's one of the worst ways to do fraud. It's like, here's my $20 Chinese food.

A (142:59)

Yeah, yeah, yeah. Okay, so let's think about going deeper in this analysis. How else could we unpack this? One way I'm thinking is, well, meals make a ton of sense. If you're working late, you expense a meal. But also there's just the normal business activity that like you might sign up for a new marketing software on a Tuesday. You might also sign up for it on Saturday. If you're doing that on Saturday, that's probably even less likely to be fraudulent because why are you putting your credit card down with our latest sponsor TurboPuffer? Because, you know, if you're signing up for a database tool and you're putting your credit card down now, maybe they run the credit cards at a different time, who knows? But I'm wondering if there's a, if there's a layer to the onion that you could pull back there. Are there any other data points that you think you'd be looking at in the next few weeks that might help you crystallize this and really, really steal up your position for the, for the haters.

C (144:04)

So the point you're making is really good because there's someone else who really interesting reply I got was someone who works, I guess in credit risk. And one of the biggest signals they had about whether or not an application for their finance software product was fraudulent was whether or not the application was submitted on a weekend. And they started looking into all of the ones that they got on weekends. And most of them were false positives. At least the ones from sf.

A (144:28)

Totally anecdotal information from sf. Oh, because people. Because it's fraud everywhere else. But in SF it's the grind.

B (144:35)

I mean this is really bad for everyone but SF because it shows that the great lock in is really only happening in San Francisco.

A (144:43)

It's local area.

B (144:44)

It's a local phenomenon. It's not actually people in New York. Oh, great lock in. Great lock in.

A (144:49)

Okay, so going forward. Yeah, we gotta look at if business software purchases are up in San Francisco on the weekend. I also want to know if it's going to spread, if we're going to have some memetic contagion and we're going to see the great lock in potentially spread to New York. That would be likely for me. I want to know Miami, Austin might.

B (145:08)

Need to get a little bit colder in New York for potentially.

A (145:11)

Potentially.

C (145:13)

Anyway, New York got a little bit close to it, but not maybe a quarter of the effect size.

A (145:18)

Wow. Huge.

C (145:19)

And it was only after 8pm it wasn't like in our chart for SF. It really is starting at about 9am 10am wow.

A (145:28)

What about timing? You said this is a new phenomenon in San Francisco. If you scroll back the timeline. When do you start seeing a takeoff of this 996 culture in San Francisco.

C (145:40)

Over the last six months.

A (145:42)

Six months. It's that recent? Wow. Yeah. That's fantastic.

C (145:46)

We don't see this in 2024 really? At all.

A (145:48)

That's crazy. I feel like 2024 was definitely a year where people were working, but I guess it just picked up a whole new, whole new gear.

B (145:57)

That was. The great capital wars began.

A (145:59)

Yeah.

C (146:00)

When we had this idea, I really didn't expect to see anything because when you. It's pretty rare to see these kinds of spend shifts in any demographic.

A (146:09)

Yeah.

C (146:09)

Especially like when I looked at it nationally, it was completely unmoved because most people don't really change something as worn in as their eating habits that often even you know, tech employees or workers. So I really, I really didn't expect to see anything. And the only time you would expect to see something like this. This is a pre to post pandemic trend. 2019-2021. A bunch of public and private data sets started to show shifts in where and when people were eating.

A (146:38)

Right.

C (146:39)

A really classic one from the pandemic era was movement from outside the city center. So you know, people were spending much less time downtowns. And then suburbs start seeing a lot more activity. That was a really popular one during the pandemic period. But after the pandemic you really see stopped seeing this kind of shifting. Most people kind of set in their ways and things were back to normal. So the fact that we saw this, only saw it in SF and that it was so recent is pretty rare and shocking.

A (147:07)

Last question for me on the 996 question. Part of it is not just the 6, not just Saturday, but the 9 to 9. Are you also seeing or do you think it's worthwhile to look into what's happening from 5pm to 9pm because that's also the hallmark of the nine. Nine is that it's not like nine to five, six days a week, it's nine to nine. And so you should expect more employees expensing food late night on a Tuesday, on a Wednesday, on a Thursday, on a Friday because they're working longer hours. And so is there a way that we can measure the weekday hours worked by San Francisco employees?

C (147:52)

Yeah, one of the really great data sets for this would probably be GitHub commits.

A (147:56)

Oh, GitHub commits. Yeah, we got to call up. We had the CEO of GitHub on. We should ask him to pull something. Now of course there is a public data set, but I think that's mostly for open source projects. And I don't know if the. I know that there's like a heat map of when people commit, but it's only by day. I don't think it's by time. So the data would probably be internal, but I don't know. I mean, it's a Redmond Washington company, so I don't know if they're going to put something out, publish it that says that San Francisco's outworking Redmond. You know, that might be a little controversial for Microsoft to do. Anyway, thank you so much for hoping for joining. Anything else to share? We're good.

C (148:40)

Thanks guys for having me. Subscribe to my substack econlab.substack.com econlab.substack.Com.

A (148:49)

Thank you so much for hopping on the stream. We'll talk to you.

B (148:51)

Thanks.

A (148:51)

Have a great rest of your day.

B (148:52)

Cheers.

A (148:53)

Bye.

B (148:55)

Next up, we got Zach from Warp coming in.

A (149:00)

Before we bring them in, let's tell you about Wander Find your happy place. Book a Wander with inspiring views, hotel graded menus, dreamy beds, top tier cleaning, and 247 concierge.

B (149:08)

That reminds me, I actually need to.

A (149:10)

Go wander myself, but better writing it.

B (149:12)

Down in my notes.

A (149:13)

Oh, yeah?

B (149:13)

Yeah.

A (149:14)

Okay, well, we have our next guest. Maybe. Yes. Okay, great. Let's bring them in.

B (149:22)

There we go.

A (149:26)

How you doing?

B (149:29)

This is Warp. This is Warp Country.

A (149:31)

Warp Country. I don't understand why I'm wearing a Warp themed cowboy hat, but maybe I'll explain what's new in your world.

F (149:41)

So I've been filming a western. Basically you can see me doing a product demo on a horse.

A (149:48)

Very cool.

F (149:48)

Because why not? We just did a major launch called Code country where the theme is all about coding on Warp. And we decided it'd be fun to do it with a little bit of like, hey, let's show how Warp can help you wrangle your agents. So that is why you're wearing a cowboy hat.

B (150:06)

Zach, I'm pissed off because I was with John last Friday before we, literally before we got to the office and we saw these fine hats. And do you remember when I was pitching you on doing hat as a merch for company that we work with? And I was the tagline for the campaign was like I was going to, I was going to pitch this to Vanta.

A (150:27)

Yeah.

B (150:27)

You say like, this is Vanta Country, America's vampire. You beat us to it and you did it very well. So. So it's a great concept. You need to, you need to create things that, that people will remember you by. And I don't think people are gonna forget the three of us sitting here with these hats on just talking about Code Country.

A (150:50)

Yeah. Okay. Explain to me what it means to wrangle agents. I feel like I've used agents pretty reliably. Like I just. This weekend I need a new breakfast place. I kicked off ChatGPT agent mode. It went off and hunted around. I got some weird ones. I posted that. I asked for a table that was circular and it found me a used wooden spool that's used to wrap coils around. And it's actually for like 50 bucks, probably the best circular, round, big table you can get. So it was thinking outside the box. It was great. And then I've also used Claude code. Obviously I think of that as agentic. We've talked to other folks on the show, but I saw this interesting dashboard on X where someone had some visualization of all the different agents that they'd spawned out. What does the frontier of using multi agents look like?

B (151:48)

What does the frontier look like?

A (151:50)

Look these days, around these parts, the.

F (151:54)

The situation is such. If you look at, if you look at pro developers using agents in their daily workflow, it's. They kind of produce stuff that's not shippable a lot of the time.

B (152:08)

They produce security. Hey, you can't argue they do produce security vulnerabilities there. You know, it's like, yeah, that you.

F (152:15)

Can ship if you're not careful. You can ship security holes.

B (152:17)

You can.

F (152:18)

You get to a point where developers on your team don't even know what they're coding. It's really a mess to vibe code in production right now. And the thing that we've been focusing on is like, how do you get a tighter feedback loop where you can see what an agent is doing, you can review its work as it goes. You can make sure the people on your team who are building with agents are like, you know, comprehending what they're doing, that they're code reviewing the agent's code in the app. And so this is a real problem. If you look at the stack overflow, latest stack overflow survey, like number one problem that developers have developing with agents is like they produce hard to debug code that they don't understand. And so you kind of wasting a bunch of time. So yeah, we're just trying to help, help people keep their agents, you know, steer them as we talk, as we say.

A (153:09)

I want to do this whole thing. Yeah. So what do you think about these here Vibe code cleanup specialists, people that are making a living now going around.

B (153:20)

Going to the frontier.

A (153:22)

Going to the frontier and cleaning up old Vibe code and protests.

F (153:25)

You gotta bring sheriff into town sometimes, clean up some messy, some messy.

A (153:33)

Fallen planets.

F (153:34)

That, that to me is like a symptom of something gone wrong with the tooling. If you're getting to a point where you need like a janitor or whatever to come in, clean up, clean up your code. You want, you want to get the engineers on your team to be able to produce shit that actually goes out to production.

A (153:51)

I guess my other take on this bad code cleanup specialist that we're seeing pop up is like, I don't know, there's a world where there's the type of person, like the ideas guy who's like, I have an idea for an app and I want someone to build it for me. Then there's the type of person like you, Jordy, who you have an idea, but then you're a Figma expert and you're going to work on design and you're going to turn over something that's really communicative in terms of the interaction design, then there's one layer further. Now with Figma make and other Vibe coding solutions where you could actually turn over a full prototype that's pretty functional, but it's not really going to scale. And so maybe the Vibe code cleanup specialist is just saying like, hey, I'm a really talented engineer. I am truly an engineer. I know computer science, but I'm willing to work with the. I work well with the type of people that express themselves and what they want to build as Vibe code. And maybe that's an interesting pattern as opposed to the software engineer who likes to work with somebody who turns over a PSD file or FIGMA file.

E (154:57)

I think if you look at it.

F (154:58)

Like that, it makes a lot more sense to me. It's like, what's the new version of Mocks or prd? It's like this working app, but it's not a shippable app because it's like the agents kind of can't do it yet. It's also where you run into problems, is like these things are amazing for the 0 to 1 use case. They're less good for the like I'm working at some big company that has millions of lines of code and I want to add a feature to it. And so for that you really, you know, it's like it needs to go. You can start with like the Vibe coded prototype, but you're going to need someone to get in there who actually stands behind what they're sending up to their teammate to review.

A (155:37)

Yep, yep.

B (155:38)

I hate to get personal partner, but can you talk revenue? I think you guys have a pretty revenue ramp camp.

F (155:46)

Yeah, we're doing awesome. So we're, we're adding million ARR every like seven, eight days, something like that.

B (155:55)

Which is.

A (155:57)

You don't see numbers like that around these parts very often.

F (156:00)

Certainly this is like it's, you know we're not not. It's not a ghost town. It's a little bit of a gold.

B (156:06)

R. It's gold rush.

F (156:10)

But it's pretty exciting times. Like it's, it's super fun. The vision that we have of like building you know a development tool which is for the ground up for like how do you go all the way from prompt to production. It isn't like your run of the mill ide or like the 20th just like text based cli app. It is a unique tool built to get you know pro code out is really resonating. It's super exciting fun. Time to be working on it.

A (156:40)

Well, thank you so much for taking the time to stop by the old TBPN saloon.

B (156:45)

The saloon, yeah.

A (156:46)

We'll talk to you soon. Have a great rest of your day.

B (156:49)

Great to catch up.

F (156:51)

Thanks a lot for having me on.

A (156:53)

See ya.

B (156:54)

That was fun. Code country.

A (156:56)

Another Cohen coming into the studio breaking down big fundraise. Very exciting.

B (157:02)

I'm excited for this one.

A (157:04)

Legendary poster. One of the greatest to ever do it. Very excited to have have him on the show. Andrew Reed's in the chat.

B (157:13)

What's happening?

A (157:14)

We need to give Andrew Reed more context.

B (157:16)

We put these on not for you, although you are from Texas, but. Oh yeah, you're Warp, the CEO of Warp. But we'll leave them on because they're fun.

I (157:26)

I thought you guys put them on for me. I was gonna ask that. That was my first question.

B (157:31)

No, no. But, but, but, but yeah. Do you ever wear a cowboy hat around the office?

I (157:36)

I do not. I have boots. I'm not a hat.

B (157:39)

Okay, there you go. Boots. Boots are. Boots are strong. What's happening? It's great to great to finally have you on the show.

I (157:46)

Thank you guys for having me. I needed something newsworthy enough to make it on here so I had to raise around just to get on the show.

B (157:53)

Well, I think you've been on many times via your post.

A (157:56)

We have reacted to many of your.

B (157:58)

Posts long time, long time guests. But what? Yeah, give us the whole kind of of bet. I mean since it's your first time on the show in person, give us kind of quick history of the company and the news.

I (158:11)

Quick history is we got started last April after a team of us had been at Carbon Health for three and a half years, left in January, started the company in April, raised around, got to work and then fast forward 16 months later. We just closed our Series A and it's been a fun time.

A (158:29)

So.

I (158:33)

Wow, look at that. I thought you walked.

A (158:34)

I thought you walked away.

I (158:35)

You're like this. I'm out.

A (158:38)

We got a gong. We got a gong. Congratulations. How are you positioning right now?

B (158:42)

Yeah, yeah. The main thing that I'm trying to understand from like, like your, your journey, like makes total sense, like being deep in the weeds and, and understanding what, what you know from your time at Carbon, it feels like this category, like so many. I'm sure you're getting sick of like refreshing like TechCrunch and seeing another like Voice AI company that, that, that is going after the market broadly. But like how is the, what's the shape of the market? How is it evolving? Like where you got, what is, what does your GTM look like and all that good stuff.

I (159:15)

Yeah, I mean, here's my general take on the space is like you've got a group of us who are specifically focused on healthcare AI or like, you know, whatever voice, sms, conversational AI for healthcare. You kind of have to be specialized on it in order to make it work. And I think that's why, you know, we saw Sierra just raise at 10 billion. You've got all the call center companies with their own flavor of like, we're no longer shitty IVRs. We like have a conversational AI now. And so problem is that as you like start to hear about how their implementations are going, you start to hear how they go from like cross different verticals. They're just not doing that good of a job at it because they don't understand sort of the nuance, just crazy stuff that happens in these healthcare clinics. And so we, we literally have to like almost go on site with customers and understand their workflows, their appointment triaging reasons, like how you actually get a patient to the right place and then read and write back into their systems of record, which many of them, you know, unlike. Again, I'll use Sierra as the example, but like unlike Sierra, being able to just integrate with Shopify, who has great APIs to go do, like, hey, what's my tracking number? What's your return policy? Can you check the status of my order? We have way harder to work with systems of record and I think a lot of folks just don't want to go put in the work or understand how to go make those things happen. And so I think you'll see companies like us, be a lot more successful in getting these practices live. I think you'll see some of the larger companies maybe win some deals and then go spend nine months building custom software to go make it work. And then I think you'll see most people pivot out of healthcare over the next couple years because just because it's conversational and generative and OpenAI exists doesn't make the work any easier to do right now.

A (160:50)

And that's kind of like the return to the way the industry has always been. I feel like whenever you dig into healthcare, it's like, oh, well, they're not using. They're not using the standard SaaS product for, you know, it's like the whole genre of vertical SaaS is the textbook example where maybe they're using different ERP, different payroll, different help desk software, different CRM. Like. Like they've always had their own little area carved out. And I don't know how much of that is just like HIPAA compliance. It's like, it's certainly a unique thing.

B (161:23)

What's the gold standard now? Like, what are you guys trying to deliver on the product side? Because I saw a screenshot you shared of Sierra on there. You were throwing a little shade on the day of their launch. It was an exchange where you're just like, can you find me some green shorts? And they were like, well, we have lots of clothes. Why don't you look around? That was pretty funny. But what are you guys trying to deliver and what do you think best in class is right now? And do the models even need to get better for you to deliver on your vision, or are they good enough that it's just more about deep integration and workflows?

I (161:58)

Yeah, I think that there's a lot of tension in someone like a Sierra who's got a very almost deterministic style chatbot, and that's been around forever. I mean, what they're doing is not new. They're just saying, hey, it uses is generative AI versus what you used to do, which is keyword matching and then sharing some response that you had queued up that was approved by the customer and the chat experience. And then on the whole other side, you have an experience like ChatGPT or Cloud, which is purely generative. Very, very general purpose. Anyone can chat with it and you can really get it to do whatever you want it to do within, as long as their guardrails don't pick it up. And somewhere in the middle is a conversational experience where you call in and you say, hey, I'm having this, this issue. I have this, I need an appointment. Like here's my symptoms. And the agent has a defined scope of work, but it still needs to follow somewhat of a deterministic path to get you to the outcome. Right. So in order to schedule, I have to look up your account, verify who you are, find the location that you're looking for, find available slots, find available providers, and then ultimately book the appointment and write that back to the practice management system. So there is a workflow or a step that has like steps that have to happen in some sort of order to do that. Right. End to end is very complex. A year ago, even when we got started, you really couldn't do scheduling end to end unless you were building what was a phone tree of like, you're in this step right now. Choose one of these three options. Now you're in this step. And that's not what I think anyone wants to be the gold standard for an experience when you call into your provider. And we've always been fully conversational from day one. And I think over the summer we made a bunch of breakthroughs on how we build a whole suite of agents that work together to get the job done. And so for example, if you're calling into one of our practices now and you say, hey, I'm making an appointment for. I'll use like Med Spa as an example for Botox. There is a Botox agent that all it knows how to do is schedule Botox, but it's still conversational. And then if you're like, hey, I actually need to switch to a laser hair removal appointment, then it switches to the laser hair agent. The patient never sees anything. But in the background we're doing a bunch of multi agent switching. And that's the only way to make these things work, work and still be generative. And they still have failure rates like 5 to 10% of cases. It doesn't work as expected. You can imagine all the dumb that patients say it's like completely random. And so. And you can't plan for all of those scenarios, but you just have to get. This is why I don't think AI doctors will be a thing anytime soon, because that margin of error will always exist. And it's like they don't want to take liability or responsibility for that. We can get away with a little bit more margin of error because we're doing administrative functions, but that's the experience we want it to be. As you call in, you know that it's AI, but you're having a conversation with it. Just like if you were talking to ChatGPT's voice model and it can do the job end to end, even if it's scoped to a limited set of jobs.

A (164:38)

Yeah. How much of how you build this is like, develop an RL environment to actually post train a model to interact with all the back office and administrative systems versus kind of like create your own deterministic SaaS layer that then your agents are kind of interacting with or are just kind of like function calls within deterministic, like business logic.

I (165:09)

It's very much the latter. It's like we've got this app layer where it understands an ontology of a practice. Services, locations, providers, all of those have their own specific customers context. Right. There is different protocol. If you're scheduling one appointment versus the other, maybe you're like three steps into a triaging workflow now and then. They do have tool calls where they get to say like, I need to create an appointment, I need to find available slots, I need to create the patient account. All of those tool calls have their own business logic and agents that they interact with on their own. It's all very constrained. So you kind of have like 70% as prompt engineering, 30% is code in the background to be like, hey, that's not the phone number the patient patient's calling in from. You can't look up that patient. Right. Because there's security factors to consider. So definitely a lot more of the latter than it is like any sort of RL environment where we're like, you don't really. The conversations don't change that much. So we rely on the foundational models to handle a lot of the edge scenarios where someone's trying to go off rail and say something that's not in scope of the agent.

A (166:10)

That makes a ton of sense. Thank you.

B (166:11)

Do you. Last question. Do you guys internally view yourselves as building vertical software with the end user experiencing the product as an agent? Because I feel like there's been a lot of companies come out over the last year that saying we're building AI agents for xyz and then if you actually drill down into, okay, what are you doing and what is the experience of when you. Your customers? It's like SaaS. Right. And that's like, it can be AI enabled.

A (166:42)

To be clear, we're extremely bullish about that.

B (166:45)

Yeah, we're. And we're su. And yeah, I'm super bullish about it. But I. But the way that you're described, the way that you're Describing this and the companies that, that are, that I think are, are sort of doing good work broadly are like, it's not like you're trying to sell that. You're just completely reinventing like an entire business model. It's more like, like we're building really valuable software and automated workflows for companies. And the end result is that the. Is the patient or the user will have a great experience.

A (167:13)

Sell the solution, not the technology. Yeah, and I feel like one of the first companies not, you know, who knows, but maybe one of the first companies kind of embrace the new stance.

B (167:22)

You don't have a AI domain, for example.

I (167:25)

No, I don't think the practices really give a shit that it's AI versus some other system. Yeah, they, they care. Like we alter. I've said this, I've said a few things from day one. But like, one is that there is inherent product market fit in what we're doing as long as we can make the agents work as expected. And so if I can schedule appointments reliably, if I can refill prescriptions, if I can do all those things and the front desk doesn't have to do it, or the call center doesn't have to do it, like, everyone will buy that if it's cheaper, faster, better, you know, whatever, you name it. The second is that, yeah, they just like, don't care that it's AI. They think it's interesting. What's nice is like every group has an AI strategy. They don't know what that means yet, but they know they want to partner with a company to go do that. And so we treat. It is all partnerships right now because it's so early, but we're definitely building vertical services, I would say that use software to make it really efficient and you don't have to go hire people now to go do the work in most cases. But yeah, it doesn't matter that. If it were, if it were, I suppose an IVR that got like a phone tree that was able to do the same thing, like someone would buy that too. They would just get replaced by the thing that's better and has higher conversion, better patient experience longer term. And so, yeah, that's really the philosophy. And then longer term we are going deeper into the stack in terms of the software that we will offer to the clinics to also help them do their jobs better when it's not the AI doing the job. And so we look at it as like right now we're doing very specific bespoke services for the group, like answering calls and scheduling or doing outbound campaigns and texting a bunch of patients saying, hey, you're overdue for your annual wellness visit or your vaccines or whatever it is. And then having a conversation and again getting them scheduled longer term. You can imagine, like we build more of the software that sits again on top of their practice management system, but that helps the actual clinic teams do their jobs more efficiently as well.

B (169:17)

It's just.

I (169:17)

Just deeply integrated with the AI component.

A (169:20)

Fantastic. Thank you so much for taking the time.

B (169:22)

Great to have you on. Join anytime.

I (169:24)

Great questions.

A (169:25)

We'd love to have you.

I (169:25)

Thank you guys.

A (169:26)

We'll talk to you soon.

I (169:27)

Text me whenever.

A (169:28)

Cheers. We will later. See you. Up Next, we have 11 labs coming in the studio. He's in the restream waiting room, but not for much longer. Let's bring in from 11 labs.

B (169:44)

Welcome to the show.

A (169:45)

How you doing?

J (169:47)

Hey, John. Hey, Jordy. Thanks for having me on.

A (169:50)

How do you say your name? Because I almost greeted you with a hoy, matey.

J (169:54)

It's Mati Stanischevsky. Although I had the same problem with Jordy. So I was texting people, is it Horny or is it Jordy?

A (170:03)

We're gonna Spanish roots there. That is fantastic.

B (170:07)

Oh, yeah, yeah. If he went hardcore Spanish, it might be Horti Orti.

A (170:12)

Yeah. Anyway, let's jump right into it. Give us the news. What's the latest in your world?

J (170:18)

The Latest is today Eleven Labs is launching a tender offer of $100 million where we buy early stage employees and investors at a 6.6 billion valuation, which is double. Thank you, guys.

A (170:34)

This is amazing. This is amazing.

J (170:37)

So. No, so happy to be able to offer that to all the believers from the early days. And we're building the company for the generation. So I'm hoping to align everybody on that build for that journey.

B (170:50)

So somebody came to you, they wanted to give you 100 million. You said, I don't need it in the business. How did this come about? Because is somewhat non traditional to. For a company to launch, potentially hire.

A (171:04)

One more AI scientist for 100 million.

B (171:06)

Right? Well, yeah, a lot of companies would. Maybe they'd say let's raise a couple hundred million and we'll do some secondary as well. But what kind of.

A (171:16)

Yeah. What put you in the position to just focus on a secondary transaction in this round?

J (171:23)

Yeah, two things. One, we are growing very healthily as a business. We got to 200 million in IR are over the last weeks, which has. Thank you. Thank you. It has been a interesting journey where we launched the product. So we started the company 2022, launched the first product beginning of 2023 it took us 20 months to get to 100 million in revenue, which is at a time super quick. Now there are some even quicker transitions. Then took us 10 months to get to 200 million in revenue, which is where we are today. And hopefully we'll get to 300 million by end of the year in five months. So 2010, five months. So we are growing very quickly and we are doing that in a healthy manner. We have money in the bank to invest in GPUs in international expansion, so we don't really need much more capital in the bank. And then the second piece is.

A (172:17)

We.

J (172:17)

Had a combination of acquisition attempts over last month and of course as someone who came back, turn them down. Before 11 labs I was at Palantir, my co founder was at Google Piotr and we had a very different setup. Palantir at the time was private. It was almost 15 years private. The equity wasn't very liquid. And with Piotr we want to make it very different. Now as you think about 11 up, so offer frequent liquidity to employees. If we continue running in a healthy manner, we are having the capital ready to deploy. We want to make sure that the employees can think about this next five, 10 years. So that's where that early liquidity is really helpful. And you are right, we had amazing partners with Sequoia Iconic joining in, keen to lead the round, invest more in the business and we of course love working with them, so wanted to to give them more at the table as well.

B (173:18)

Give us a breakdown of where all the revenue is coming from. Obviously I can think of millions of use cases and I've seen different companies leveraging 11 labs. But where are kind of the pockets where most of the growth is coming from?

J (173:33)

Yeah. So at eleven Labs we have two key parts of our business. One is a creative platform side of the business and now the second one is our agents platform. Most of people will have known us from that creative side. That's something that has come with when we started a company that's of course voiceovers for movies, dubbing of movies to other languages, narrations for audiobooks, creating music, bringing music into the fold. And here that's over now. Few million monthly active users that will come through the platform and create incredible content. And then a second part. And here we've seen an incredible adoption over the last two years where enterprises like Cisco, Twilio, Epic Games, bringing the experiences into Fortnite are building just incredibly new voice agencies. And here you can think about call centers, customer support, personal agents, massive media. And across the business, roughly we are approaching 50, 50 between the two. So on one side, the self serve, the creators are half of the revenue and then enterprises are the additional half. But that kind of other side is going quicker. Where we've seen.

B (174:48)

Sorry to interrupt. Are you competing with the Sierras, the Fins, the Decagons, or drilling down into that enterprise business? It's also, it sounds like, I imagine if somebody's creating a video game, they're leveraging voice agents as well. But could you say a little bit more?

J (175:10)

Of course. So on one side we power a lot of the companies. A lot of the companies you mentioned are our clients. Decagon, Maven are some of the example places we work directly. So a good example is Perplexity who will use voice as part of the interaction to use Perplexity. Another is Epic Games where they've effectively deployed Darth Vader experience in Fortnite so every player could interact with Darth Vader life for the first time. One of the biggest deployments. As a Star wars fan, I can do nothing less. And then of course so many across other use cases. Chess.com works with us where you can have personalized learning experiences when you play chess. And then other companies like Cisco Twilio, where they do use it for both internal use cases but also bring it to their clients in those cases. It's similar to what you mentioned, Jordi, at the beginning, where we power their work internally but also for their clients. So today we both, we power a lot of clients across and then also work directly with some of the holistic voice operations.

A (176:18)

Okay, help me understand a little bit more about where the business fits. Fits in. I feel like you've trained actual models. You're a foundation model company in many ways. And then you're also an inference seller and you have clients that effectively buy tokens from you. They buy MP3s or WAV files probably, but you're doing the inference for them. So you're kind of full stack in this AI world. And then also I just feel like it's been awesome to see you go on this run because there were probably a lot of haters. I feel like I saw some haters being like they're going to get steamrolled by the other laps and you have it and you. So what are the sources of defensibility and why like. Yeah, yeah. Why do you just keep.

B (177:10)

Why do the haters keep being wrong?

A (177:12)

Yeah, I mean it is, it is, it is it's been an amazing run and it's such a unique position that I think it was contrarian, it was overlooked early on and people didn't see what you saw. And so kind of walk me through a little bit of the structure of the business and where the defensibility is coming from.

J (177:31)

Yeah, two answers there. I think the first piece is so across 11 labs we started a company with my co founder Piotr is an incredible researcher to assemble. We think are some of the best researchers in the world in audio say we combine the kind of the foundational research, build a horizontal infrastructure and then have vertical applications on top of that. And I think the main reason that we've been able to stay ahead is clear focus on voice. We always build the models in voice. We build applications around voice and that's something that on both sides on the product side and the research side, many of the companies just aren't doing. Whether it was text to speech initially and bringing more emotional model, then creating a speech to text model that was heavily beating OpenAI, Google and others on benchmarks and now the full orchestration across those models and finally bringing music into the fold. I think a combination of just raw incredible research talent that is now at 11 labs missionary focus of building those models and then being able to deploy them in production quicker has been helpful on that side. Then on the other side we really care about building the best voice application. So spend a lot of time with our both on creator side but also on the enterprise side and build backwards from there. So that's the first answer but the second answer. I recently got to meet Jensen, Jensen Huang from Nvidia in UK and of course I came over, I wanted to to say hi and said that and then he responded that oh, I was an early 11 Labs user and still am but from the day you launched I thought that all the voice models will commoditize. But one thing I didn't appreciate is that the true voice experience takes an artist and you guys all are artists and that is is partly true. I think there's so much nuance across voice you need to create such a distinct experience when it comes to different voices, different languages, different dialects. So Jensen's take that a lot of our engineers, a lot of our company is being effectively crafting those artistic experiences is true. And yeah we will probably be now Nvidia client for life too.

A (179:54)

This is a great take. It's very similar to why we saw that news that Meta has a partnership with Midjourney. Meta has all the Money in the world. They have a ton of scientists, but maybe they don't have the artist that is David Hulz at midjourney, who just brings taste to those images. And so they gotta pay them, which I love to see. I want your reaction to speaking of that. Where are we in the uncanny valley of voice generation? There was news today in the Wall street journal that OpenAI is making a a full length feature animated film. And in this article they said the production, they're investing $30 million in this. It'll be a full, full feature length film. They're trying to do it in nine months, which seems totally doable in the age of AI, honestly. But they said the production team plans to cast human actors for character voices and hire artists to draw sketches. And I saw that and I just thought like, like I've been under the assumption that voice was solved. Even if OpenAI doesn't have the frontier model, if this is a demo of AI broadly, they should be able to figure this out. Why do you think that they're going with human voices here? And do you think there's something that's on the frontier? Give me your timelines for when we might see 11 labs technology in a Hollywood film.

J (181:20)

I think first of all, I think you can create an amazing experience already and happily we did. So maybe there's a model or skill help where we can work with OpenAI and help them out on that side. No, but they have amazing researchers. I think they are going to craft those experiences too. I think that Epic Games example is. It's a great way of showing that it is possible. In that case, we worked with recreating, working with the Estate, the voice of James Earl Jones. And it did sound exactly like Darth Vader. And players were almost amazed at how is it possible where you have dynamically generated script, effectively a live NPC in the game for the first time. Similarly, in our case, we worked with legends of the past like Richard Feynman, his voice where you can really immerse yourself in the story, but also with people like Arianna Huffington or first lady Melania Trump on creating their audiobooks. And the whole delivery of that experience is as good as real thing. And the crazy thing, it also allows you to do things you could never do before. You can bring those content pieces into other languages and still hear that voice. I think Hollywood is a trickier piece where it might take a slightly longer time aligning the likeness component of how you work with the people, how they get compensated and how you bring those experiences on Screen. We do hope that the first dubbed movies will get into production next year where you have even better thing than what was previously possible. You get that original expression available in dubbed movies and in 2027, hopefully the first Hollywood movie hits on the screens in English too.

A (183:10)

That'd be very cool. Thank you so much for joining the stream.

B (183:14)

Congrats, congrats to the whole team.

A (183:16)

Fantastic progress all over the place. Just. Yeah. Remarkable to see. I love that take about the artistry that goes into crafting these super differentiated house high, high tier models. It's, it's wonderful to see you. Love to see it.

J (183:34)

Thank you so much. I'm also so proud and happy of our MIT team of, of artists across in spirit. So thanks for, thanks so much for having me on guys.

A (183:43)

Yeah. Have a great rest of your day. The whole team talk to you soon.

B (183:45)

Cheers.

A (183:46)

Bye, thanks. Bye. Up next we have Dan from Truck Smarter. He's a restream waiting room.

B (183:53)

It's time to truck.

A (183:54)

People were saying are we going to truck dumber? No, no, we're going to Truck Smarter and we will welcome Dan out of the restream. He's in founder mode. Dan, how you doing?

B (184:05)

What's going on?

A (184:05)

Good to see you. Good.

D (184:06)

What's going on guys?

A (184:07)

Good to see you. Too much.

B (184:08)

We're trucking.

A (184:09)

We're locked in. The great lock in has begun. Many financings have been announced. We hear you have some news to share with us. Why don't you introduce yourself the company and then the news. Yeah.

D (184:20)

It's been an absolutely packed day for fundraising so I appreciate you guys getting us in. I'm happy to bring the party to freight. My name is Dan, co founder CEO of Truck Smarter. We're super excited to announce our fundraising and the launch of our product dispatch as well.

A (184:35)

What was the fundraise? Give me the details. What happened?

D (184:39)

Yeah, so we, we just raised from Socio Ventures which is backed by Cox Enterprises and then backed by all of our familiar friends, Founders Fund Thrive Capital A16Z BCV to continue building the absolute best tools for chunky companies.

A (184:55)

And how much did you raise?

D (184:57)

16 million.

A (185:00)

Congratulations.

B (185:01)

There we go.

A (185:02)

Fantastic.

B (185:02)

There we go.

A (185:04)

Warm up on that one.

B (185:05)

Give us, give us a quick history of the company. When did you start trucking Smarter?

D (185:12)

So we started in 2021. We first started by building a load. So for folks, for folks that are not familiar, a lot of trucking companies today. So there's what, almost a million truck drivers out there. 95% of them. 95% of these trucking companies own less than five trucks. And the way they find jobs is very similar to you going on Craigslist to buy something. Right. So all this information is everywhere. The information is not entirely verified. The people are not even verified. You're kind of just scraping the barrel constantly spending hours every single day. So that's kind of been the very core of our company. We started by building a load board that helped them find those jobs in a more efficient way. And then we started layering on a ton of financial services, getting them paid, getting them access to credit. If you think about these are the small businesses that just do not get access to a lot of these things. And because we're at the very beginning of their revenue journey, we can just see all this information. And then obviously now with AI this year, we, we, we. There's just a tremendous opportunity to just make finding a job so much more efficient. And that's kind of what we've been heads down on for the last six months.

A (186:17)

There's a poster behind you. Load boards should be free. Explain to us how exactly does the industry work? You have to pay to find a trucking job. How does that work? It is.

D (186:30)

It is absolutely crazy.

A (186:31)

Right?

D (186:31)

So if you think about it, for you to access this Craigslist type of thing, you got to probably pay anywhere from 50 bucks, 250, 300 bucks a month. At the very beginning of our company, we made it extremely clear, so we can't really go back from here. Yeah, we said load boards should be free. We built software that kind of connected all the disparate pieces of information into a single place, give it all away for free. And that's kind of the engine for our entire business. We continue to grow tremendously from that, and that's where we continue to layer on all these different services.

A (187:03)

What's the, what's the kind of competitive dynamic in the market? Like, are you going to force everyone to go free? It seems really obvious to, like, if there's a barrier to getting a customer, drop that to zero, and then figure out how to monetize the customer somewhere else. I'm surprised no one's done this before. So congrats on the innovation. But it seems like the competitive dynamic will be like, people will try and fast follow you. So is it just a race to build more financial products, better products behind the scenes? Like, what, what's the, what's the long term? Or is it just like, you got to get big, you got to get all the data on there.

D (187:34)

You got to get big. Right. So I Think for us, right, it's a, it's a race to layer on additional services. Is a lot of the existing companies. There's a lot of big load boards out there. They've been around for 50, 60 years. Right. So you think about Craigslist, right? A lot of people have tried to verticalize all the various categories of Craigslist. They've done so in a very real way. But Craigslist is still a monster. They still do hundreds of millions of dollars of revenue. So there's still a lot of this pattern with a lot of the existing companies. But for us, we're just able to layer on all these other things. That just makes our platform continue to be more attractive.

A (188:03)

Have you done what the Tribe podcast? This is a funny. Have you done it? Yes, yes. Yeah, I've done this podcast randomly. They were talking about nicotine pouches and the guy wanted to have me on. It was really fun. It's actually a extremely well run show. But in that vein, I imagine you can't go on every week. You got to reach truckers. What's working on the growth side. Yeah.

D (188:28)

So I think for us, right, we're the number one load board on the platform on the market today.

A (188:31)

Right.

D (188:32)

You search load board, you search free load board or number one or number two.

A (188:34)

Okay.

D (188:35)

That's the strategy. Right. Making it free just makes it incredibly easy for anybody to access.

A (188:40)

Sure.

D (188:41)

And as a result, you're getting all the people that are coming into the market.

A (188:44)

Right. So SEO play. Are you doing any like brand building, high top of funnel, you know, awareness. With truckers, are there specific campaigns? It's so hard because they like, like the truckers aren't in all one place by definition. Right. They're all over the country. So it's not like you can buy one billboard and you're good. You know, how do you reach people and actually create like brand awareness?

D (189:08)

Yeah, I think for us it's kind of continuing to solve that number one acute pain point.

A (189:13)

Right.

D (189:13)

Which is finding a job. And that's kind of what we've been focused on from day one. And as a result, I think we just have one of the best growth engines that can reach these trucking companies. Because almost certainly every trucking company has probably downloaded a load board and there's a very high chance that it's ours. And then as a result, we kind of have that information to just continue to offer them more services.

A (189:34)

Yep.

B (189:36)

Fantastic.

A (189:36)

Congratulations. This is fantastic. We will, we'll talk to you soon. We'll let you get back to your day. Have a great day.

B (189:42)

Load board should be free.

D (189:43)

Appreciate it, guys.

A (189:44)

Thanks guys for dropping on.

B (189:47)

Well, on that note, I got to get on with.

A (189:50)

Yeah. It is past it's 2:20. While we keep doing these three and a half hour streams, we keep saying.

B (189:55)

We'Re gonna do two and a half.

A (189:57)

We always belong.

B (189:58)

We're addicted.

A (189:59)

Tyler, any other breaking news? Tearing up the timeline. Anything you got for me? No.

B (190:04)

No crazy news?

A (190:05)

No crazy news. I don't know. Okay, that's good. Well, we will see you tomorrow. We will be live from Y Combinator demo day. Tune in. We're gonna have a lot of fun there. And then on Wednesday, we'll be in New York City. So we are going a little bit of a world tour, a little bit of an American tour.

B (190:20)

That's right.

A (190:21)

This week. And we will see you tomorrow.

B (190:24)

Can't wait.

A (190:25)

Goodbye.

B (190:26)

Have a great evening.