Leopold's 13F, Data Center Fixes, Shein Buys Everlane | Mike Isaac, Rowan Trollope, Dean Leitersdorf, Joanna Stern

You're watching—

You're watching TVBN.

Today's Monday, May 18th, 2026. We are live from the TVBN Ultra Room, the Temple of Technology, the Fortress of Finance, the Capital of Capital. Massive day today. Tons of big stories. 5 big stories I want to go through. Obviously, the first one is that the US jury finds OpenAI CEO Sam Altman not liable to Elon Musk for straying from Charitable mission because Musk waited too long to sue. Weird like technicality, I guess, but good news for OpenAI. Judge confirms verdict and that Musk's lawsuit is dismissed. We're having Mike Isaac from the New York Times join the show in just a few moments. I see multiple journalists on the horizon. When is Masi joining? Around 11:45 today. That'll be fun to hear about the story. From the ground because he went to the—

Yeah, apparently they deliberated for about 90 minutes.

90 minutes.

And they didn't really make any type of statement other than a statute of limitations.

And so Max Zeff over at Wired says jury unanimously rules that Musk's claims are dismissed on the timeliness issue. He filed the lawsuit too late. Court affirms it will uphold the jury's decision. It's over. Musk loses the lawsuit against OpenAI and Mike Isaac, the Rat King, says unanimous verdict in Musk versus OpenAI is in after only 90 minutes of deliberation. So did they deliberate today? They showed up at 9 and went from 9 to 10:30 and then delivered the verdict. Is that what we think happened? Because Friday's off in the jury, right? No Friday.

Yeah. Jury showed up this morning.

Okay.

Talked, talked for 90 minutes.

But they got to think about it all weekend and Friday. Interesting.

Of course.

Yeah.

It's a full-time job.

I guess it's just an interesting Interesting dynamic because you think you'd want everything really fresh. You'd go into it on Thursday night or something like that. Rat King says, "Huge day, wow." And what did Tyler post? He posted a video of Drake talking about something. What's going on over here? Let's play this clip.

W's in the shot.

Shot.

W's in the shot.

Is that the song of the OpenAI Slack right now?

I think that's when he is gambling in front of them.

It is a funny way to pronounce chat, but I enjoy it. Anyway, the big news that was going on all weekend, actually, there was a lot of anticipation for Leopold Aschenbrenner Situational Awareness hedge fund to drop the 13F. It was supposed to go out Friday night, 5:00 PM. Everyone was saying, oh, if he's not releasing it—

Well, people were expecting it throughout the entire day.

Yeah. They were very excited.

And then there was some speculation that he'd been able to petition to not have to release it.

That was one theory.

That was one theory. The other theory is that he was just entirely in cash.

Yeah, don't need to report it. Just wind it down.

Said it was a good run.

Yeah.

It's over.

Yeah. Yeah. He's like, I counted the ooms, and there's none left to count. We're done. Pack it up. No, quite the opposite. Leopold Aschenbrenner, the hedge fund's chief investment officer, is known for making extremely successful investments based on his core assumption that Frontier AI will continue to improve at half an order of magnitude, 0.5 ohms per year, which translates into a thesis that AI will create unprecedented demand for compute and its associated bottlenecks.

John, they're saying it is blindingly light. It is brighter here, isn't it?

Yeah, I think we got some new lights. We're sort of, you know, tweaking things. I do like that the wide is less dark. There's been a number of times when we've gone and watched videos and we've been very dark in the front. So we're bringing some light around.

We'll see.

Maybe, maybe we overdid it. Maybe we'll dial it back. I need to brush my hair. My hair's a little scruffy today. I also need a haircut, but we'll get to that.

We'll get to that later in the show. John will be getting a haircut live on the program, potentially. But before we go any further, yeah, I, uh, uh, Nick, over the weekend, oh yeah, a little gift for our very own Tyler. Tyler, let's go! Open it on the video, on the video, Nick. Waited in line. He waited in line.

Look at this.

Waited in line.

Hey, what do we got for Tyler?

Waited in a very long line just for you, Tyler, because— What is it? I'm trying to open it.

Okay. Little anticlimactic.

Wow.

It's a—

what is it? I don't know how to pronounce this. Am I reading upside down? It's a little, little watch. Let's go.

Another watch for Tyler.

It is nice. I don't know if you thought it might be, it might've been something else like the Swatch AP collaboration.

But really like the whole, you know, everything in the Swatch portfolio is fantastic, including this. I don't know. Describe what's on there. What is on there?

Yeah, Nick, what is it?

It has a rotating bezel.

He says it has a rotating bezel. Okay. But just to be clear, it's not the Royal Pop.

Which was completely sold out and causing like stampedes all over the country, all over the world. I saw footage. I think from an international country, around people really mobbing it. You were mentioning that you thought it was maybe an aura loss for both companies because of the craziness.

I think it is.

Yeah, I just—

Your brand is now associated with chaos. That's not good.

Yeah, yeah.

Right? And AP, although it's exclusive, like you have to sort of wait in line. The waiting in line is like, here, have a Diet Coke and sit in this private room while I tell you that you will not be getting an allocation in the skeletonized AP—

Come back soon.

Royal Oak or whatever, right? Come back soon. And it's a very highbrow waiting in line.

Yeah.

And this was sort of—

Yeah, they had to come out over time and say, these are not going to be limited. We're selling them a lot. And so the people that waited in line just to sell on the secondary market, I think, have done pretty well. Oh, really? At least in the short term. But I would expect that over time, prices will sort of retrace toward— retail. I did see a funny graphic of somebody that was like basically saying like, you know, comparing like getting a job versus waiting in the line to get it. And you actually did quite a bit better if you just got a job on Monday instead of getting in the line. And then over time, you know, your earnings really ramp out.

Yep.

But anyways, sorry, Tyler, if you thought that was a royal pop. I don't know why you would.

Oh, he's doing the Kevin O'Leary Mr. Wonderful 2 watches, one on each wrist. Looking good.

There you go.

Anyways, I think that could be a good, good daily for you. Who knows? It's got a little character to it. You make it your own.

Yeah, it looks good.

Sometimes the man makes a watch.

Somebody should make a string that you can turn it into a Royal, like a Royal Pop, you know, like, oh, like a lanyard type of thing.

Okay, okay. Yeah, that's possible. 3D printing. Plenty of, plenty of opportunities. Well, let's go back to Leopold Aschenbrenner and his 13F, the infamous 13F. There's a lot of discussion around it on the timeline. Really, like, we have not seen this level of attention on a hedge fund's filings in a very long time. It's because it's breaking out of FinTwit. It's breaking into Tech Teapot and Tech X and all of that. Mostly because a lot of the discussion centers around the filing shows he's made some massive puts across the semiconductor sector, $2 billion on SMH, the VanEck Semiconductor ETF. And so there's, it feels like maybe more of a pointed thesis, less broad, hey, semiconductors are gonna do well, more, I actually, me, Leopold in this case, understand where the real value is, what companies within the semiconductor industry are undervalued, which ones are actually going to be useful in the next iteration of the buildout. And a lot of stuff has been priced very hotly. Some stuff is overheated. The Nvidia trade for a while became like crushingly obvious, and then it grew so much that that was not one of his early positions. Now it is looking like he is going long Nvidia, which is interesting in the backdrop of is Nvidia a car? Do they still have a moat? Well, there might still be something else going on there. You have to dig in through this and understand what's going on. But the filing is hard to interpret cleanly because a 13F is only a snapshot of holdings as of March 31st, 2026. These positions are stale. He might have rotated out of these, meaning these positions were in place during the early phase of the Iran war. It also doesn't include Private State International.

Copy trading senators. Tends to work pretty well. They tend to be— maybe they're very knowledgeable on some of the subjects that they're trading on, some of the companies. But they tend to take a more longer-term thematic view. Whereas Leopold, he's operating a hedge fund. You don't really know. His holdings could be wildly different just weeks or days. After the end date of this.

There is the team behind the Nancy Pelosi stock tracker, stock ticker, I forget what it's called. They have one for Leopold now, although of course, it's based on the 13F. It's a loose— it probably has massive tracking error, but it's directionally on theme, like their interpretation of what Leopold would do if he was managing.

Very accurate for 3 months ago.

Maybe. Yeah. So reminder, 13Fs do disclose put and call options. They don't disclose the strike prices, expirations, premiums paid, hedge ratio, short position swaps, or whether the options are part of broader structures. So you have to be careful out there if you're trying to read the tea leaves too precisely. You can only— take away so much from these. So, uh, Feijiao— I don't know how to pronounce that— says, uh, unfathomably bad takes around this morning and a good reminder of why 13F digging is mostly a waste of time. March 31st, uh, we were in the heat of the Iran war. Makes sense to put on hedges at the time. Uh, options exposure on 13F gets quoted notionally, so as if it were 100 delta i.e., all 100 shares per contract. So, when you see something like, oh, he owns $1 billion of Intel, it's usually he owns the right to purchase $1 billion of Intel, and he has actually deployed far less capital into that position, although it is sometimes an important sign of things to come. We have no way of knowing whether these were 5 delta convexity hedges and represented a fraction of what people are saying were billion in puts or whether they were ITM puts, in-the-money puts. Further, outright shorts don't get reported either. Too much noise associated with the things that happened back in March that aren't relevant now. We have no idea about his turnover in assets and trade frequency. A lot happened in the months of April and May. His positioning could be completely different. Making investment decisions for 80 vol assets based on data from months ago sounds like a good way to burn money. So, don't idolize people and develop your own thesis for why you own and sell things. That is a good takeaway from an account that I can't pronounce but has good takes. Now, there were a bunch of funny memes about this. LeapTrader says, now drop Leopold Aschenbrenner's portfolio where he sold all his holdings and went full cash. That certainly would roil the market. I do wonder, is the market actually moving on the 13F? Are we seeing like when a position is disclosed, there's a, there's a pop of copy trading going on, or is this just sort of like online fun and games for the tech folks. Do you know?

Yeah, I mean, if you want energy, T1 Energy is up on the news. 17% today.

And we talked to the CEO of that company, right? T1 Energy is building solar panels in America.

Yes.

That is very exciting.

Chinese company that had to divest and turned into T1 Energy. And yeah, I think we first talked about T1 in Q4 of last year. Done very well since then.

I'm excited about it. We can bridge into that in just a second, but Investor Nick says, "Did that leprosy fella tank the market with his 13F?" 47 likes, but very, very funny to just massively mispronounce Leopold's name. Anyway, where should we go from here? Options on 13F, everyone repeat after me. Citrini is reminding everyone that options are reported with notional value. So be careful out there. The interesting bridge is just around the AI backlash and the fact that a lot of—

Situation in the chat says Bloom Energy is actually down.

Well, he's owned Bloom Energy for years, or maybe not years, but like 3 or 4 13Fs have disclosed Bloom Energy and that one has been fully digested. By the copy traders, I imagine. Anyway, the AI backlash is continuing in a bunch of different ways. And one interesting sort of twist on this is that a lot of the AI maxis, the AI bulls, were sort of concerned at least that this would all be fossil fuel-based build-out because everything else was too slow. They might be— they may be fans of nuclear, they might be fans of of solar, but it was seen as infeasible, seen as the timelines being far too long. So if Leopold is in fact taking a position in T1 Energy, that sort of leads me to think that there's a little bit of a shorter timeline to at least bringing some solar power to bear during the AI buildout, that it's not all just sort of, a hope and a dream that there will be solar power on the grid in any near amount of time. A lot of the nuclear power companies are moving on the backs of the AI buildout, but it's still 2032, you know, when we talk to these folks. Even the optimistic ones. So there has been big pushback on AI data centers across the board. We've talked about this a bunch, and it's both a left and right wing issue now. Sagar Engetti predicted this, I think last year when he joined our show, and it's, It's been interesting. Left wing is worried about job displacement, theft of art, destruction of creativity. Right wing sees them as surveillance centers. That's the latest term is that they're used to spy on people. So that's an anti-libertarian, anti-right wing position. But there are a whole bunch of others. Just, you know, this hollowed out coal town is voting right wing and then data center comes to town and they see it as, you know, just making their town worse off and benefiting like the coastal elites and like the—

People have flagged too that both sides are using AI to create graphics to oppose data centers.

That's true. Yeah, there's all these like deep ironies. There was a whole piece on someone who's protesting data centers and using a lot of AI to research how she can push back.

Gabe says data centers need to be rebranded to data ranch.

Data ranch. I like a data ranch. That's a good one. Anyway.

We got aux-powered.

Oh, interesting. Salty says that Leopold sold Blue Energy in the latest 13F.

Trimmed.

Or trimmed. So if that's the case, then there you go. And yes, Lulu does have a good breakdown of the narrative mishap, which we can go through. But the latest debate that I saw was over this huge data center in Utah that's being championed by Shark Tank's Mr. Wonderful, Kevin O'Leary. Are you familiar with this whole thing? There's some renderings. It actually looks really cool, but it's weird because it's like, I see this like beautiful glass building, I'm like, "It's not gonna look like that. There's just no way. There's no point. Like, why would they ever build that?" But someone dug into—

We're in the render economy.

Yeah, someone dug into like the plan, and the plan actually seems pretty reasonable. But Mr. Wonderful, he's sort of an over-the-top caricature of a businessman. Like, he plays one on TV. He is a real businessman, but he also plays a businessman on TV. And so he's a bit of a soft target. Like, he was recently seen sporting not one but two expensive watches, not unlike Tyler Cosgrove over there. Uh, he went to the Oscars wearing a Cartier Crash Skeleton and a ruby Rolex Daytona. And I believe he also had a, uh, like a trading card around his neck. So very ostentatious, very over the top, a very soft target if you're looking for someone to target in like a, he's doing it for the money, you know? Like it's pretty easy. And so if you wanna paint data center construction as maybe not in the best interest of average Americans, Kevin Leary is gonna do a lot of the heavy lifting for you.

Mr. Wonderful in the context of developing large-scale infrastructure that people are afraid of. Yeah. Sounds like a supervillain too.

Yes, and also like we can put this in contrast to Eric Schmidt or Tim Cook where the previous generation, like the major hyperscalers, like the big tech companies, they've done a pretty good job building a lot of infrastructure, making really, really bold climate pledges saying we're gonna be net zero by this year. Our data centers are really clean. They built a lot of data centers without really any disruption. There was no backlash to Google Cloud. Through 15 years or 10 years of building AWS. And so now—

But neither of them were rocking dual iced-out—

So you're making the case for quiet luxury, the quiet luxury of a Tim Cook or an Eric Schmidt, potentially.

Definitely.

Yeah, I mean, in this case, it seems like—

No, I think Mr. Wonderful is not the guy to be the face of—

Potentially not, but apparently his actual data center plans are reasonable. It actually seems pretty by the book according to current plans. It's in a remote area. It uses its own power and water, and it doesn't seem to disrupt any local communities. We can pull up this video from Quick Thoughts that has a little bit of a breakdown and goes through, I think it's called, I think Quick Thoughts calls it, Why I'm Not Opposed to the Utah Data Center. I think the big Utah data center is, combine. So this is a 4-minute video, but we can, we can watch this and break down.

Do you have a link?

Quick thoughts, his thinking. It's in the timeline. Um, because there was a, a TikToker that was reacting to how bad it was, and he is saying it's actually not that bad. So let's play this clip.

Million views complaining about a giant data center in Utah, and I'm kind of confused by that because I would think that an uninhabited desert valley in Utah is the perfect place to build a giant data center.

I've been following really closely what's happening in Box Elder County, Utah, where Canadian billionaire Kevin O'Leary—

who's Canadian, that's a big—

data center, a $100 billion project. Okay, this would be the largest data center in the world at over 40,000 acres, and at full capacity, the data center, which is called the Stratos Project, is set to use 9 gigawatts of electricity.

Gigabytes, you saw that?

Double the entire amount of Electricity used by—

well, she said, she said it correctly.

The data center is built. Yeah, yeah, but, but the transcript said gigabytes, which is funny.

It's not—

AI fails again. We need another data center to fix that.

Steve in the X chat says TVPN Studio uses the equivalent of 23 atomic bombs of energy to produce niche technology content.

So large is because they are buying water rights of the current property owners. So the current property owners are using water for agricultural irrigation. The data center project buys that land, buys a huge—

He makes this sound good, but then it's like, wait, are we going to have less food? That doesn't seem that good. But the point is, is that he's not taking it from like someone who is going to be paying water or some local community. It's like there's, there's, there's already water rights there that are staying in that valley.

It's not drawing power from the grid. If we look at electricity consumption by state, we can see that Utah just doesn't use that much electricity compared to other states. There are plenty of states that use double or triple. Tennessee is about triple. Pennsylvania, 4 times. Texas is like 10 times, more than 10 times what Utah uses. So if over the course of this project they reach their goal and they double or triple Utah's electricity usage, So why is that bad? It's not incurring more cost to the people of Utah because they're building their own power plant.

By Utah as a whole, Robert Davies, a physics professor from Utah State University, says that he actually thinks the project will require an additional 7 to 8 gigawatts of waste heat energy, meaning that the project in total will be 23 gigawatts of total thermal energy, which is the equivalent of dropping 23 atom bombs in Utah every single day. Also, let's—

okay, electricity generation across every state is going to have that same thermal load property. Not every generator is perfectly efficient, so they're going to generate waste heat as well. So if you say, okay, we're going to have 23 atom bombs a day worth of electricity going off in Utah, well then, currently we have 230 atom bombs a day going off in Texas.

You got to put everything in an atom bomb comparison, like Your car is like the size of like 5 atom bombs. Like an atom bomb is like maybe this big, maybe a little bit bigger.

Yeah.

Your car weighs as much as 7 atom bombs. That makes it sound so much more like weighty when you're like just comparing everything to atom bombs.

Night temperature by 28 degrees.

This is actually pretty crazy. 28 degrees feels like a lot.

Daytime temperature could increase 2 to 5 degrees throughout Hansel Valley. not the state of Utah, the valley where the data center is being built. Same with nighttime temperature, could increase up to 28 degrees trapped in the valley. Hansel Valley is an uninhabited desert valley. So if you build a big power plant here and a big data center here, maybe it'll increase the temperature of this valley by 5 degrees. But okay, nobody lives there. I think this project solves a lot of people's stated concerns. With data centers. You're worried about water usage? They're reallocating agricultural water to cool the data center. Worried about power cost?

They're building their own power. That one line is not helping.

Worried about waste heat?

It's helping his—

But I like vegetables.

In the middle of an uninhabited desert valley where it's already hot. And you're worried about this is such a huge project? This is a giant data center or something? World's biggest data center? Well, that's just data centers that don't have to to be built in other places that are being built in this uninhabited desert valley. I think the concerns in her video are just fearmongering for reasons that I hope I've explained here. Thanks for your time.

I guess the question is, like, they say that there's water for agricultural usage right now in that valley, but that valley's uninhabited and seems like a desert, so it doesn't seem like they're growing food there. So like, where is that water actually going? Because is it just getting piped to to some other farm, like far away? Or was it like they were planning this farm?

So way, way, way back in the day, way back in the day, you could just have a piece of land, you could drill a well, and you could pull up as much water as you wanted. And then people realized that you might, if you have a property here and there's property here, here, here, here, here, they're oftentimes all pulling from the same aquifer. So you, all of a sudden, if you come in, you move in next to me and you start pumping, billions of gallons. Yeah, you're drinking my milkshake, right? And so it's very possible that all these parcels of land which they collectively bought, they all have their own water rights. That doesn't mean they're being used, right? So, 'cause people will sell their water rights to like a neighboring property that is.

Yeah.

And so—

But my question is like, it sounds like they sold the water rights previously, or they had some sort of deal to sell them. Send the water that they were getting out of the desert, which I can't imagine produces that much water, but I guess it does, use it for agricultural purposes. Like, what were they growing?

Well, agricultural could mean you have some cattle. There's a bunch of different potential meanings for that. It doesn't mean you're growing fresh produce.

But were they actively using it, or were they just like, nah, I don't know.

No, that's the other thing. That's the other thing, too. It could have been agricultural land. But not—

It could have been like a failed farm that's not farming anymore, like a former livestock farm, something like that. But I don't know. I feel like people are gonna want to go a click deeper on that. Like, he rebuts a lot of the good rhetoric, but there's still like another layer there to go back.

Dave says the water could be used at Amangiri.

Yes.

Influencers are protesting. In the flats outside of Amman, Jordan.

You drain the pool at Amman, Jordan, it's gonna be a big protest for sure.

It's gonna be over.

Well, yeah, I mean, these points, like as you said, I think are gonna be hard to break through just because AI is so deeply unpopular for a variety of reasons. And we should watch the video of Eric Schmidt getting booed on stage at University of Arizona Alex Kantrowitz played a video here. I don't know if we need to watch all of this, but he says this is incredible. Artificial intelligence getting booed out of the stadium in any commencement speech it's mentioned in. Maybe telling college students AI was taking their jobs wasn't the best strategy. Let's watch this clip.

The architects of artificial intelligence.

Interesting. The question is whether you will help shape artificial intelligence. We do not know.

We do not know the precise contours of what this—

if you'd let me make this point, please. Step 1: if you're giving commands, you got to bring a soundboard. You gotta be like, "AI, yeah, it's not that bad, but also I hear you." The perspective of the immigrants Who has so often been the person who came to this country. They're really going crazy.

We thought that we were adding stones to a cathedral of knowledge that humanity had been constructing for centuries. There's just a low-level boo the whole time.

It's so rowdy. Like, normally you'd think there'd be like a little bit of boo and then they just like get quiet down. Okay, this is about to turn into a riot. This is crazy. Did he just bail on this thing? No.

Your agency.

We have only seen— At this point, I mean, you got to go off script.

You could—

you can't stay in a script.

It is funny that if you cut it up in the right way, you could make it seem sound like the most evil. It's like, you will surrender your agency.

Yeah. Okay. Now we need to take this clip. Do that thing where we—

Where he says he's lucky they didn't flashbang it.

We need to do that thing where we take out the boos and just leave his words and then add cheers.

So it's just the same exact speech, but everyone's just like, "Yes, this is amazing!" I can try to find it, but there's a video of him after the speech getting mobbed by students. They're all yelling at him.

Really?

They were not fans.

Wow. This is rough, rough, rough. Yeah, not good. I mean, the big thing is like, I don't know that that is, like everyone is booing for a slightly different reason, but it's like this ensemble of problems and grievances with AI generally. Like everyone is, one thing that I've been like frustrated about is everyone is vibe coding like 24/7, leaving MacBooks open, talking about like productivity. And yet the like the magical moments the consumer technology has been like completely left behind. Like there was a time when we got the cloud, we were building a lot of data centers, but every year you'd get like a cool new thing, like Yelp would come out and it was like, it wasn't changing the world, but it was like, oh, you could find a cool new restaurant and maybe like, or Groupon, like Groupon was like not a great business ultimately, but like for the first couple months of Groupon, you could like go try a restaurant for like half price and it just felt like, magical, or like Uber, when that came out, it was like, wait, I can go out and the car will be right outside instead of having to like call a phone, call a taxi cab service, maybe it comes, maybe it doesn't, stand outside in the cold, try and flag a car.

There were all these things. Wait, do you think they were, I'm just thinking that, do you think they were like angry at usage, nano banana usage limit?

Probably, probably.

Is that, is this whole thing just a misunderstanding?

They might think we're in a plateau. and they might just be upset with the lack of progress outside of coding domains. They say, "Yeah, the writing is just still not that good on any of these models." I can clock it. I can clock it. Yeah, it's still clockable.

Yeah. Yeah, at first I thought they were mad that like at Google, Eric Schmidt, he was doing too many stock buybacks instead of investing in technology innovation. Yeah, too much cash on the balance sheet.

Yeah, having $100 billion on the balance sheet in cash is just unacceptable. Like, yes, you get Waymo, yes, you get DeepMind.

Yeah, 'cause it just says they don't know what to do with the money.

Yeah, yeah, they weren't innovating for a long time, and then that makes a lot of sense. Makes sense why you would boo them, sort of the Tealian.

The Tealian boo.

The Tealian boo. No, and then also like, yeah, the jobs thing is super real, like whether or not AI is affecting the job prospects.

It's also, so we should pull up Lulu's critique, 'cause I'm sure it'll be way better than this, but just in those handful of sentences, Like, is that, that felt like a speech more potentially like oriented towards maybe like the Stanford student body, which is like, how are you going to contribute to AI? That's what I was like, sort of, that's what was standing out to me. Being like, don't be afraid of this thing, like jump in and help shape it. And if you're maybe someone in Stanford, and you have the opportunity to go actually be involved and you're at the epicenter of all this progress, maybe that would land. But at U of A where people are hearing like, hey, all the different career paths that I'm thinking about.

I would prefer, in terms of commencement speaker, I would prefer someone like a Sam Sulek to give the commencement speech. That would be like my, like Eric Schmidt is like, he's kind of like a, Meh, Sam Sulek, that's an inspirational speaker. That's gonna fire me.

He's on the come up, right?

Exactly, yeah. Did you have a question?

Derek, more plates, more dates.

That would be fantastic too.

Yeah, I was trying to, Gabe's asking about the, why would he give a speech there? I was trying to find a connection. I think he's just a big name.

Okay.

And it's very, obviously his experience is very relevant in this moment.

Show up tomorrow, none of you are getting any jobs. Just terrible. Yeah, no, no, I mean, there is this thing where like, AI needs to create jobs because like even if AI isn't destroying the jobs, if we have a weak economy, there won't be good jobs. And then like you're still hold, you're still held accountable for that. And so you gotta create jobs. And then on the data center side, like there's just so many issues within that that we can go through, environmental impacts, which are probably real. If you burn a bunch of fossil fuels, you're gonna have negative externalities, diesel generators, these things are smoky, the air quality, all of this stuff is, is fairly real when done improperly, which is happening. The water use thing, mostly fake, but still needs to actually be walked through fully and digested by the public. The noise issue, which is solvable, but still not that great. And then a bunch of other issues that are just not gonna happen magically. Ben Thompson had a wild, wild proposal. He had a great piece, which, I wish we had time to read through the whole thing, but we can sort of run through it. So he starts with an anecdote from Politico. A Texas county southwest of Dallas this week passed what may be the state's first county-level moratorium on data centers. Not what everyone was expecting in the free state of Texas. Everything's bigger in Texas except for the data centers, which are getting smaller now that there is a county-level moratorium seeking to buy time for lawmakers to soften the blow of development sweeping across rural areas.

What's the county?

Hill County's commissioner, Hill County's commissioner court voted 3 to 2 Tuesday to put a year-long moratorium on data center and power plant construction in unincorporated areas, citing an influx of as many as 8 data centers planned there, many of which could have their own power plants. Opposition to data centers is spreading in regions led by both Democrats and Republicans as politicians try to balance economic development.

Yeah, so apparently according to— According to AI, there's no official public count of operating data centers in Hill County, but there's 8 proposed or planned data centers. So this is a place that—

They're gonna be delayed.

Yeah.

In Missouri, one small town unhappy over its city council's approval of data centers voted last month to oust all 4 incumbents running for reelection. In North Carolina, Governor Josh Stein has made a point of saying that sales tax exemptions for data centers cost the state up to $57 million per year. Texas has hundreds of data center locations operating or in development, second only to Virginia among US states. The growth has stirred pushback from environmentalists and rural residents who worry about the effect on water supply the electric grid, or their quality of life. Officials in states across the country are starting to have second thoughts about data centers, and some are looking to roll back tax incentives. And Ben Thompson says, "I chose this story because it happened just, it happened to have happened over the weekend." In truth, there are an exploding number of options, including one just up the highway from where Ben Thompson lives in Wisconsin, in DeForest. "And they're hardly isolated sentiments. 7 in 10 Americans oppose constructing data centers for artificial intelligence in their local areas, including nearly half, 48%, who are strongly opposed. Barely a quarter favor these projects, with 7% stronger in favor." Now, I was thinking about what do Americans want to build? Because it's easy to look at the data center stuff and be like, well, everyone's against building data centers. But I do think that there's an element of like, Americans don't wanna build anything. Like, I was reflecting on the whole reindustrialization meme this weekend. I got a version of that sweater mailed to me that I picked up, and I was thinking about the actual knock-on effects of reindustrialization. Like, most people don't want a car factory in their town.

But we do want new roads. Well, not necessarily new roads.

No. People don't want new roads and they don't even want the roads paved 'cause they're like, "Ah, I'll just buy a bigger car." Like, I don't know.

Hospital?

You want a bunch of people dying next to you? I don't think people want hospitals. I literally, golf courses, they have poisons, they're bad for your health. Like, I actually think people just don't really want change necessarily. They don't want things built broadly. Like data centers are probably at the bottom of the list. Like they're the least popular, but they're like high-speed rail. I thought that would be popular. It was not popular. And like, I'm just going down the list of like, oh, like you want, like, oh, we need, maybe you're a national defense person. You want a missile factory next to you blowing up bombs? Like, no, no one wants that. Like, what do we want? Like, we don't really want anything. We're kind of good on building in America. I don't know, I just think we're good. Like, we're just like, we're fine, it's good. Don't change anything. No new trends, no new growth.

Yeah, I think when there's self-interest, right? When people wanna build their house, right? When people wanna—

Everyone wants to build their data center. Yeah, their data center for sure. But people don't want other stuff built generally. Like there are very, very few things that people are like, yeah, I'd be down for that to be built. People like the status quo. They're happy with things as they are and they don't like change. So like anything new is going to be like, somewhat unpopular as nuclear power was. Not building out nuclear power 50 years ago was of course one of the greatest mistakes humanity has made and one that contributes directly to data center opposition today, given questions about the impact on energy bills. Also interesting, we have to do this another time, but the, you know, did we run out of nuclear scientists? Was that what stopped the buildup? Did we not have enough geniuses? I don't know, maybe we'll dig into it. But Ben Thompson has an interesting solution. He points out a bunch of ways to fix the problems of data center construction and opposition. He says first—

People are saying homes in the chat, but then again, people don't really want more homes in their area once they already own a home.

They block 'em all the time. They block home construction all the time and also permitting and also expansion of existing homes. Like these things, I'm not saying that they're like, as unpopular as data centers, no way. Data centers are at the bottom, but homes are something, maybe in the abstract, but like new housing in communities is like razor's edge, 50/50, 60/40. Like there is a lot of opposition to building just in America broadly. Like that's just the nature of our society. So Ben Thompson has some solutions though. What do you gotta do to build a data center properly? He says, first, this sounds obvious, Tech needs to fix its messaging problem. The issue, and if an answer seems obvious, then there surely must be some other problem at play, is threefold. First, a good number of people in tech, particularly at one of the leading labs, genuinely believe most jobs are going away. They could lie more effectively, but beyond being dishonest, it's also a betrayal of the fanatical devotion with which they are pursuing AI despite obstacles, including the challenge of spending billions and billions of dollars on models that are obsolete in months, if not weeks. Second, it is extremely hard to describe the benefits of inventions not yet made, cures not yet discovered, economic activity not yet engaged in, et cetera. This is always the burden of those arguing in favor of progress, and the sheer potential of AI actually makes the problem even harder. 50 years ago, everyone was like, "Electricity isn't that expensive. Why do we need to build nuclear power plants? They're scary." And now electricity's expensive, and we're like, "Oh, we should have built those." That's the way these things always go. Third, tech is and always has been terrible at understanding and relating to the rest of society. I go back to how Silicon Valley was extremely skeptical of Facebook, a company predicated on connecting with friends and family precisely because it's filled with people running away from their friends and family. You can optimistically say that people in tech live in the future. You can also more cynically say they live in opposition to and denial of humanity for better, and in this case, for worse. Second, tech could control the misinformation. TikTok is a major point of this. He talks about how the algorithm is still controlled by the Chinese, and maybe there's misinformation there. Second, in a rather ironic twist, Meta has learned the lesson of trying to control misinformation, doesn't wanna overtly censor, but now the company gets no credit for not censoring misinformation about data centers. And so it's like this weird thing. And then third, this was a wild card, which I didn't think of, but X is the social media platform X and Twitter, formerly Twitter, is actually incentivized to be anti-data center in a weird way because X is owned by SpaceX. And a big part of SpaceX's upcoming public offering is the possibility of building data centers in space. This is like total tinfoil hat, I think, But it's an interesting like, okay, and he says, to be clear, he hasn't seen any evidence of thumb on the scale or not, I certainly haven't. But, you know, part of the problem though is that we would never know if there were. And so he goes on to propose something very, very bold, very, very bold. He says, instead, the most obvious solution is the most crass, simply start giving people money, not universal basic income though, If data centers are a resource for our AI future, then start paying people for that resource. If that data center up the road weren't sold to my neighbors based on amorphous tax benefits that my local government may or may not spend appropriately, and I was talking to Tyler about this earlier, but rather were to result in a check in the mailbox every year, I suspect you could get a lot of people on board. So he put some numbers together and he says for the data center up the road, it was expected to be 1.6 gigawatts, which could generate around $3 billion in annual operator revenue. De Forest, the village it was to be built in, has around 11,500 people. So you could pay every person in that village $10,000 a year, and it would only equate to 3.8% of annual revenue grossed by the data center. And he says, I bet that that proposal would've been approved And I bet the operator could very easily pass on those costs to actual data center users. It also highlights how relatively pathetic the original commitment that I think the data center said, "Hey, we'll give you $50 million," which is like nowhere near what that math works out to. So data centers coming to town, you get to vote for it, but the data center company says, "Hey, we'd like you to vote for this and we will give you a $10,000 check in the mail." every year forever while we're operating this. And that seems like that could actually get people on board.

So, yeah, this goes back to even months ago at this point, we were saying, you know, AI is not a, is not like a, you know, natural resource where you benefit from having it in your backyard, right? If you're just an everyday AI user, you do not care where the data center is at all. And so if someone is coming to put it in your community, it's pretty fair to want to benefit from that in some way. And like a direct payment like that, I think, I'm sure that will happen more.

Yeah, yeah. And what I was talking to Tyler about was, does the, like, do local communities feel a difference between $10,000 in the mail directly to them, or $10,000 to their local government that says, "We're gonna use this to build roads and hospitals and all the different things that we do." Like, I think that on net, the average American is a little bit skeptical about dollars going to the government actually benefiting them at a 1-to-1 ratio. They definitely think that if the money that goes in is worth something, but a lot of it gets mixed around and there's delays and false hopes.

Yeah, and the data center's already gonna generate a bunch of, local tax revenues for that local government.

Show me the money. Show me the money. That's what the locals should potentially be saying.

Well, that's what I'm saying. It's like they don't, you know, the, I think it's totally fair for the local population to think, okay, like this big infrastructure project is happening in my town, even if I'm not going to work there. Yeah, it's going to generate some taxes to help improve our community. Show them, give me the money, basically.

Give me the money.

Go direct.

Yeah, go direct with the money.

I like it.

Well, we have Mike Isaac from The New York Times in the waiting room. We can come back to our data center debate after we check in with Mike. And I think he's on location. Is this correct? Mike, where are you? Welcome to the show. How are you doing?

I'm good. Can you hear me? I'm sorry. I'm literally outside. The courtroom.

Amazing. No, we can hear and see. It's fine. It's amazing. Well, take us through it.

My actual—

how has today been going? What's happened?

Uh, it was crazy. Uh, basically today was supposed to be the first day of jury deliberations, and we were a few reporters with the courtroom because in the morning it was about, uh, both sides presenting their case for remedies to the judge on basically how much money, if anything, would be dispersed as a result of the lawsuit. Literally, in the middle of this deliberation, the clerk goes and interrupts the judge and says, "Hey, da, da, da, da, da, something's happening," basically, this scurrying, and everyone's like, "Oh my God, what's happening?" Da, da, da, da, da. This is less than 2 hours into it, they reach a verdict. The jury comes back in and delivers the verdict.

Interesting.

What was your expectation going into today? Did you think you'd be hanging out at the courthouse all week?

Yeah, yeah, I'll see you soon. Sorry, that's a lead OpenAI counsel walking by that I should go run after, but he's doing his thing.

I mean, we're just hanging out.

We're just hanging out.

I'll bug him later. Okay, very literally, he was just chilling and walking out. I, uh, sorry, I can't see without these. I, uh, uh, I forgot what you— I'm so tired. What did you ask me?

I was— I was— what was your expectation for your week? Were you expecting to be at the courthouse every day?

Yeah, we were. Like, I got here again at 6 AM and like was ready for a long, like, sitting out in front of the court for days because the way these work is like you get 10 minutes notice from when the judge gets the jury verdict to get down here. I live 10 minutes away, but still, like, no reassurances. So we had me, my colleague Cade Metz, and then Natalie Roca, another colleague of mine, just like ready. And I was just like, thank God, when they came back, because I didn't want to sleep out here.

Okay, so the actual verdict, it feels like a victory on a technicality. And what I'm interested in is, that over the last few weeks, it feels like the core discussion or the, or the talking point was Elon Musk, you can't steal a charity. Very pithy phrase, easily memorizable, could stick with you or could bounce right off you. But you know what his grievance is. And then OpenAI sort of needs to say, well, the charity still exists and we had an agreement that we would go this way. And it was a little bit more complicated. But that doesn't seem like what the jury actually decided based on. And was that— like, as you think back to the last 3 weeks, do you think that there were actually good seeds planted around the statute of limitations and when the case should be filed? Because it feels like from the reporting and from the viral, you know, the screenshots and the emails and the quotes, like, there was never like, oh, yeah, we all remember the smoking gun of statute of limitations. No, I don't. I remember the "You can't steal a charity" or the Brockman diary, right? And it feels like we got a different outcome here.

I think I remember at different points, like, they— this whole debacle only became a thing after the launch of ChatGPT. And, you know, the company was showing, you know, massive traction and revenue growth. But I never heard specifically, like you said, this statute of limitations.

Yeah, but how did you process it?

Well, that's a wonkier point too, right? Like, it's, it's very easy to— and that's what I think, like, really the strategy on the Musk side was, was to go for really, like, clearly digestible talking points for a juror who may not be steeped in nonprofit contract law or statute of limitations and exactly what that is. And I think that's what they were betting on, too. They're like, all right, if we can sell the jurors on this idea that Musk is, you know, selflessly trying to, you know, interrupt something that could be bad for the world versus OpenAI's more technical point of, look, you should have filed this lawsuit years ago, maybe they can win it. And so I think that was going into it what everyone was kind of thinking about, like, is this going to be Certainly what I was thinking about is this going to be a battle of like the billionaires, who do you trust, who do you— it's like a character thing, that is this a referendum on that. And exactly what you said, it's super surprising when they came back and essentially I would say statute of limitations was like, if that was— that was the ball game, right? And if they had blown past that, if they had not find them the burden to be met, then we would have seen how it really played out. But that was just— that was the whole thing. No.

What? So, so last week I was surprised that, that Elon jumped on the China trip with, with Trump.

Oh, yeah.

Was that a blast? Yeah, that, that something. I mean, a lot of the people online were just like, he's a billionaire, he can do whatever he wants. That was like, that was—

president supersedes the federal judge. Yeah, I don't know if that's actually the case.

There was some dialogue around like, hey, you're in the middle of this historic trial, like, you should be present, or at least able to be present. Did that— do you think he did that because he felt like it wasn't going his way and he was just like, I need to make the most of my time?

I think he— so, so I, uh, yeah, NBC wrote a good story on that. Like, he was not, uh, excused. He could have been recalled and asked to testify again, uh, and Typically bad form when you leave the country to do— when that happens. Um, and, uh, so what I was told, or what I heard, is that they had actually spoken to the judge beforehand to like make sure it was like okay, and like that he probably wouldn't be recalled. I think part of it also was that both sides were— both sides are on a clock, so you only have so much time to, to present your evidence And the early testimony was running long, so OpenAI still needed to get through a lot of the testimony of their expert witnesses towards the end. So they decided— and Musk's side also decided they weren't going to recall Musk. So like there was that part of it that probably made it okay. That said, like, it's probably a bad look when you make it the first 3 days of the trial and Sam and Greg make it basically most of the time. But, uh, but at the same time, like, it didn't come down to character who pissed off the judge necessarily. It came down to like a legal technical argument, which seems to have— this jury was pretty sophisticated, at least in like focusing on something that I didn't know if it was going to land or not.

Yeah, did, uh, it— I mean, it really makes all of the, like, the the AI safety testimony feel like maybe a miscalculation because it sort of took the conversation in a completely different place and then they got focused on this like technical issue. I mean, the jury doesn't put out like a statement. Are we, are we expecting any sort of like closing statement from the judge or is this what we get here?

We— so by the way, sorry, there's There's still like people protesting in the background if you wanna see that, but on my very terrible laptop camera.

Are they protesting the statute of limitations 'cause they're on either side or?

Protest, there's actually, this has been the best part of this 'cause like there's many different protest camps and it's kind of hard to define who is against what.

Are any of the protesters protesting other protesters?

I mean, genuinely, yes, probably. Yes. There's the, I wanna do a survey on that.

Yak is out there protesting the D-cells.

Genuinely, there, there were supporters. No, 100%.

Yeah.

Um, so actually, usually post-trial, you people like me go and try to find a jury and chase them down, which is what we were doing. I think they probably are already out of the building. Um, I ran around the back and saw a van that was like all blacked out and this marshal that I had known the whole trial, and like they were just like getting the hell out of here. So I'm guessing they didn't want to get mobbed by Yeah, us. But, uh, the judge— I'm gonna try to get the notes out— the judge left the jury with like a pretty good summation, not of the trial, but just like appreciating a jury and like respecting a jury finding, like finding, uh, parties liable or not liable, you know. And, and I think that the point of that was she didn't—

she—

some federal judges could like be like, no, I'm throwing your verdict out or whatever, but she respected the jury. It was a jury of their peers, and they were deliberate, you know, and they listened intently. And so she left them— I'll find the exact quote and send it to you guys—

but she left them on sort of like a, "We thank you for your service." Yeah, yeah, that seemed also a little bit unexpected because when the jury verdict became, you know, popularized or publicized as like advisory, a lot of people were sort of interpreting that as, well, like, it doesn't matter at all in that case. But it seems like the judge did— wind up sort of, you know, giving the jury a lot of weight and very quickly reacting to the jury's verdict.

And I think that's really important as far as appeals go, because you could argue—

Bias charge, cut out.

Like, you could argue, like, oh, the judge— yeah, exactly. The judge didn't care. The jury— so I think there's real incentive to be in line.

Yeah, yeah, that's very interesting. What was the snack setup today? Are you going to get a proper lunch now? I feel like—

My God.

I feel like that was one of the most disappointing arcs, if I'm going to be completely honest with you. The lunch game just didn't seem to evolve.

You were saying that— You weren't learning from your lessons at all. You know the Nathan for You episode where he's got the chili suit?

We were going to do that for you. Because it just felt like the day— Okay, day 3, you show up with an apple and a banana. It's like, okay, he's still learning his lessons. Acid, but like fool me 7 times, I was expecting a Chipotle burrito or something with a little more substance. Get into the 4 digits of calories, please.

Oh God, people were like DMing me saying I have like scurvy or rickets by the end of this trial. I think I just have like a really disturbing diet overall. So yeah, and today I forgot, I was out last night until way late at a show and I'm hungover and I forgot to bring food. So it's just, this is, This is basically my, you get to see my slow descent into madness, but thank God we're done.

Okay, so I mean, we asked you earlier, is this the stuff of movies? Is there gonna be a movie about this or was this anticlimactic?

I think like, I think the movie's still going, man. Like this thing is still, there's so much. I feel like this is the most exciting time in AI OpenAI is really on its back foot in a lot of ways. This gives them some relief in the many fronts that they're being attacked on, whether it's going public this year with a messy balance sheet, or Anthropic coming after them, Google coming after them. Google I/O is tomorrow. So like, if anything, it's a brief reprieve, you know. But, uh, I wouldn't, I wouldn't make the movie now. I'd wait, wait a couple of years.

Okay, okay. Anything else, Jordy?

The story continues.

Story continues.

I'm expecting to see Model Ys around San Francisco that say, "I bought this after Elon lost his landmark trial against OpenAI." Yes. The bumper sticker.

New bumper sticker.

Right on.

Well, have a great rest of your day. Thank you so much for taking the time.

We'll talk to you, Mike. Great to see you.

Next time. We'll talk to you soon. So Mark Cuban has another proposal. For how to deal with data centers and internalize all those negative externalities. He says we should tax tokens federally at the provider level. Tyler, you're gonna have to interpret what this would mean in all the ways that companies would wind up getting around this with maybe less robust answers potentially. He says, not a lot, less than $0.50 per million tokens. It will accomplish 4 things at least. It will push the big AI players to optimize tokenization, caching, routing, and localization, which will reduce energy usage, saving them in energy costs more than what they paid in tax and reducing strain created by the growth in energy consumption, which will generate maybe $10 billion a year to start, but over the next 10 years could grow 30x to 100x. So he's thinking 2 orders of magnitude in a decade in terms of growth for AI. That's low end of what a lot of people think. And then 4, create a source of funding to pay down the federal debt or deploy in response to the things AI brings that we don't expect or don't like. At some point, the models will pass it on to consumers. Of course, that's okay. Consumers will have the benefit ability to choose between providers or do, or to do everything using open source models locally, which I guess wouldn't be taxed.

What do you think? This is kind of like the opposite of what we were saying before of like going direct, right? Because we were saying, okay, uh, you know, the actual data centers are gonna make so much revenue, you can just tax the data centers and then the money goes to the local community and then that's where you see the benefits. But isn't this going like up the chain even more? So you're taxing the companies, then people in the community like definitely won't it, like, the money will be, like, so abstract if it's at the federal level. I feel like this is the wrong way. Here's something else. You should be giving people— you should get a check from OpenAI and Anthropic every month, maybe. That's, I think, the better version of his, if you want to tax the companies. What if we taxed companies, you know, what if we had something like, um, like a sales tax, or, you know, what if— Profit tax on profit.

Income tax.

Yeah, like, if someone— when someone is paid— What if some of that money went to the government to help pay for public services? And maybe even if a company is doing really, really well, then you could take a percentage of their profits.

Yeah.

Because that company has to—

And if investors sold their stakes, they would also pay a tax on whatever gain they made on the stock.

And then every single— what about every single underlying vendor that the company— you had the same sort of like— structure for every underlying company.

So like if Nvidia sells a bunch of GPUs and they make a bunch of money, they don't pay tax on the profits on that.

Yeah, or even somebody like a contractor that manages a building, right? So they have a, maybe it's a small local business, they manage an office.

They make a million dollars, costs are only half a million, that half a million profit, that gets taxed.

Has anyone thought of that?

That might work anyway.

And then you could use that money to sort of COVID the costs of operating the government and then even potentially use some of the extra to pay down the debt.

Potentially. Well, Palmer Luckey is going back and forth with Mark Cuban about this. Palmer Luckey says, there are already massive economic incentives to optimize. So this is just a tax on American companies that makes foreign models and products more attractive, along with creating the infrastructure for government to track all AI usage and punish anyone who doesn't report. Mark Cuban says those incentives change over time. Right now the incentive is to grow and spend market share over optimization. You know this. Don't— do you think the marginal cost of some bips on a token is going to make those buyers choose differently, or do you think the models are just a commodity and price is the only differentiation? Space, and then the question mark. Every time, you know it's not AI. And the tax would only be on what providers sell, not open source models, not local, not internal and what foreign models are you referring to, Palmer? Mark, you are essentially making an argument for central planning. The burden is on you to show where it's worked before. No quotas, no mandates, just good old capitalism and competition, Palmer says. This is obviously not capitalism or competition by any reasonable definition. This is a tax that specifically disadvantages one type of AI business to the benefit of others, artificially propping up their business models. And my business is one of the ones that would benefit because he's not token heavy. That is interesting take.

Semi Analysis says 50 cents, uh, per mTOC is a lot of money. Mark, are you considering— considered cash hit on prefill or just output tokens?

These are the hard questions. Thank you.

Steven says imagine a bit tax in 1995.

Yes, flops tax. Uh, I don't know what, what else is going on in the AI slop world.

The bot farms appear every time you— how about this? What about every time you move your curser, it's just 1 cent.

Mm-hmm.

Right?

Yeah. I don't know. Tax on something.

It's pretty funny. I was saying last week when I was saying like, basically reinventing the US Postal Service.

Yeah.

A lot of people were messaging me saying, you know, people, you know, this exists already. It's like, man, it's tough when the sarcasm doesn't break through. Doesn't breakthrough.

Well, the bot farms have figured out anti-AI, anti-data center posts on Facebook are good for engagement, but ironically they're using AI slop to do it. You don't know this is the slop. This might be the most perfectly designed set of stones ever visited upon a beach. It's not worth— it's not worth giving up an inch of this to a data center. Indiana breaking: an Indiana resident, a reportedly arranged stones to make an anti-data center message. This is 99% slop. And this one is really sloppy.

Wow.

Wisconsin's forests, farms, lakes, rivers, small towns, not a single square inch of Wisconsin is worth giving up for an AI data center. Interesting that the I in IS is capitalized. Makes me think that that was added after the fact, but the rest is, Pretty sloppy, but kind of beautiful. I kind of like the perspective on this image with the big farm and the barn in the background and the—

This makes me want to visit Wisconsin.

Yeah, does Wisconsin actually look like this? If it does, perfect place to build a data center. Yeah, that's the only thing it's missing.

Well—

No, I want— yeah, we have to go and find the ugliest 10,000 acres in Wisconsin. New challenge, Tyler.

Well, we gotta cover Everlane. We gotta cover Everlane. There was a bit—

well, just to close out, should we, should we, should we cover this post from Ken Griffin that was going around?

I, I want to cover this, but like, the trick is that this is a Ken Griffin clip. So basically he pivoted on AI. 3 months ago he was saying like, uh, it's not really useful, the reports that we get from AI models. Are not actually, you know, relevant to our business. And now he's saying, for us at Citadel, it's allowed us to unleash a much broader array of use cases. It's been really interesting to watch work that we would usually do with people with master's or PhDs in finance over the course of weeks or months is being done by AI agents over the course of hours or days. And it's seen as sort of a blackpilling moment because he says like, I got home and I was sort of like, sort of, "I gotta tell you, I went home one Friday fairly depressed by this because you could see how this was gonna have such a dramatic impact on society." And it is like a weird moment and it's sort of like, "Oh, okay, he's waking up." But then, if you actually watch the full interview, this is 1 minute from a 40-minute interview or something, and he goes on to enumerate a whole bunch of different benefits and where he is allocating his workforce. And also, Citadel's in a very, interesting game theoretic dynamic where it's not, they're not a monopolist. So by definition, like they are in competition with all their other funds. And so there is a world where like, even if they're getting incredible value out of AI, they wind up using AI and humans in conjunction to compete because we are in like the centaur era, which is sort of what he alluded to. But anyway, what did you want to take away from this?

One thought I had is that Citadel has—

AI psychosis, that's what you're saying.

No, they have a team of thousands of PhD-level talent that are doing things that AI can do pretty well now. And him driving home on a Friday being depressed, Part of me was thinking, is he depressed because he realizes everyone will soon have access to 1,000 people with PhD-level talent that they can turn on? And maybe they can't cover the whole market. And obviously, you know, he talked a lot about, you know, how much software there is to build. He's like, we'll never build enough software. But at the same time, he was thinking like, wow, this, this like resource that I've accumulated, this like capability, this like this team when AI can do what they do and everyone can access AI, like, how is my business gonna change?

Yeah. So funny reflecting on the time I worked at Citadel and my job was basically to copy and paste cells in an Excel spreadsheet. And so I wrote a Visual Basic script to sort of just do it for me. And then I was able to just like have 7 hours of free time every day. And I wound up being able to do a lot of other stuff. And it was a story of automation. And I can tell you, at least back in 2011, I spent the summer over there as an intern. There was a lot of stuff you could automate for sure. A lot of stuff in the back office, middle office, and yeah, some research in the front office. But Citadel's Edge is more It's more than just research. They do a lot of CEO interviews. They talk to a lot of people off the record. They have a lot of information that does not exist on the internet.

They have scale.

Yeah, there's a lot of—

Track record.

—of different stuff. So I don't know. It's interesting.

Yeah, let's talk about Everlane. Yeah.

Well, I think we have our next guest already here. But we can go through Everlane quickly, or Or we can come back to Everlane at 12:45.

Let's do that. Okay.

Well, let's bring in Rowan from Redis because he's waiting in the waiting room. Rowan, welcome to the show. How are you doing? I'm doing great. Thanks so much for being here. Since it's your first time on the show, I've actually been—

Are you getting an act— Is this an active sauna session for you? It does look like a sauna background.

No, I'm in Tenerife today, actually.

Amazing. Amazing.

The wood paneling behind you really does look like a sauna.

No, I like it. I like it. Well, yeah, we'll see in a few minutes if you start sweating.

I'll be doing the cold plunge next.

We'll see how that goes. Yeah. Well, some people in tech do combine cold plunges with— That's happening. —talking about their companies. Tim Draper being one of them. Yeah, yeah, yeah, for sure. Anyways, great to meet you.

Great to meet you. I've used Redis a ton about a decade ago. I'm a big fan of the product. But if you could introduce yourself and the company a little bit before we go into the news today, that'd be great.

Yeah, absolutely. Thanks, guys, for having me on. It's an honor to be on. I loved your show. Amazing. Big fan. Thanks. And watch all the time. Great. Yeah, so I head up Redis, and we're one of the sort of core infrastructure components that has been around. We're one of the bigger open source projects over the last 15 years and sort of helped build out a lot of the internet infrastructure. And got a great team. We have just about, we're 1,500 people now and we're starting to see a lot of— There you go. That's great. We're starting to see a lot of traction in the AI world as people are starting to really build out agents. So as you guys were just talking about, lots of opportunity there and so we're being pulled in on agent data.

Yeah, I wanna get to that. Is the correct framing for Redis for people who might not have actually used the product in-memory key-value storage, like non-relational databases, think like MySQL, but less structured and also held in memory, therefore faster?

Totally, you nailed it. The history of it, it's an in-memory data structure server. It's not really a database. Yeah. But it's been treated as a database. And the killer app that kind of took off and made Redis a part of, kind of got the tendrils into all the applications applications on the whole internet was key-value. Yep. Used— being used for caching. Yep. So it originally started as an in-memory database. The big thing that's changed though, and this is just coming live now, is over the last few years we've re-architected Redis and launched a new product that uses flash as the backend storage. So now we have the fastest, the world's fastest flash object store. Okay. And so that's a new thing, and that was really driven by AI because we were seeing huge demand for way bigger, way more data and also RAM prices have gotten crazy. Sure. And NVMe cost performance has improved dramatically.

Sure. Okay, so then take me through some of the history of the business. I know you joined as CEO, like in the modern era, but in terms of that transition, what is the shape of the business? Because a lot of people are building open source software and I'm always fascinated by that transition and that interaction between between the product, which sometimes has like incredible developer pull, incredible ecosystem, and then also an incredible opportunity to build a real business around it, but what is the shape of that? Because I think people go to Red Hat, they go to consulting shops, they go to hosting providers, enterprise software wrapped around it, but what— how would you describe it, the shape of the business around the product right now? Yeah, so we're—

it's a great question. Open-core company. So we have an open-source base, which is Redis. Anyone can download it and use it for free. It's used all over the place for free. And then we have a paid version. So for example, the, the recent innovation I mentioned, the rewrite using Flash, uh, storage, that's, that's not for free. That's something that you would pay for. We have a lot of performance advantages in the paid version. We have a hosted version. It runs on all three of the major cloud. So if you get Redis on Amazon or Google or, or Microsoft Azure, right, we, we provide— we have our own version essentially that runs on those clouds. Um, and so that's, that's, that's the, that's the heart of the business. Most of our usage on the internet is free Redis because the free product is amazing. Yeah, the paid version is even better.

I like it. That's good, that's good salesmanship. So the, the relationship with the hyperscalers, uh, is that consumption revenue that's coming to you? I set up an AWS instance, I pull Redis off the shelf from the dashboard of a million different tools, and then as I'm using it every month, stuffing more and more data into it, that money's flowing to you from the hyperscalers?

Exactly. So it's a little different depending on which hyperscaler you're talking about. For Microsoft, they're hosted— if you buy the first-party service from Microsoft, right? So on the hyperscalers, you have first-party services that are offered by the hyperscaler themselves. Then there's third-party that you buy through the marketplace. In Microsoft's case, when you buy Redis first-party, it's actually our software. And exactly, you're exactly right. We get a revenue share of that. So that's called Azure Cache for Redis. And then there's, uh, and then Amazon and Google no longer offer as first-party services Redis. They have their own products that were once built based on Redis, but we did a license shift to kind of get them off of our tail, frankly. So Amazon and Google now have their own code bases that they have to maintain that, that, that have really diverged from what is now core Redis. We offer on Amazon and Google through the marketplace Redis, you know, as the Redis cloud product essentially. And then increasingly we're offering that through new cloud vendors, either they're neo clouds or like Vercel, for example. So if you ask an agent If you're building on Vercel and you say, hey, please deploy Redis Cloud, boom, you'll get our product. And it seamlessly is integrated into their platform as well.

Yeah, that makes a lot of sense. So, I mean, I remember when I was using Redis, I was using it a lot for actually like business intelligence and like data analysis. It was just nice to clean up some data, have it all available in memory, much faster to sort of query and do like MapReduce over. but obviously the bread and butter is caching. But I'm interested in the shape of the agent business, like what data is being stored, when, because a lot of this stuff can be loaded in context, it can live on the chip. We talked to the Cerebras founder last week. Like, there's an incredible amount of work being done really, really deep in the in the AI supply chain, and then there's everything out to hard drives and tape storage on the other side. And so what is the sweet spot that Redis is filling right now?

Yeah, so if in the past, as you just talked about, sort of the killer use case in the cloud mobile era was caching your database, basically. Okay, you could use it for a lot of other stuff as you talked about. And in the new world, we sit in a similar place, and that is essentially, providing all the context, you know, like coalescing all the context for the agent and then delivering that to the agent. Yeah. And we, we had to actually build a new product to do that. So what, what developers have been using Redis for in the agent, this like called the next era that we're heading into, is storing agent data and hosting agent context. And one of the reasons for that is that you're going to see multiple orders of magnitude more agents than human beings. In a company. And what that has a direct consequence to the load you're putting on your backend data systems. So just like in the cloud mobile era, you saw, you know, kind of you went from like, uh, guys that were sitting at green screens like bank tellers, for example, and the load factor on your backend DB2 might have been like 10,000 to 1 or something. Okay, then you added mobile and you added a million customers or 10 million customers. So, so 2, 3, orders of magnitude more load. And Redis came in there as a scaling layer. Yeah. Okay. And you didn't, you didn't have to go and scale DB2 or your mainframe or whatever. It doesn't make any sense. You Oracle that backend. Similarly, in the agent era, a similar transition is happening. So as that load increases, you can't have, like my company has 1,000 employees. I can't have 100,000 or 1 million agents. Mm-hmm. And we're going crazy with agents right now internally. Yeah. Hitting my backend data systems because I'm going to be paying a hell of a lot more to all of my underlying providers. So we use Redis in the middle as the context engine and we cache and hold all the context from the underlying databases in Redis. And that's what the agents interact with. So we launched a brand new product that's on our website right now called Iris. Yeah. And this is its exact intention is that what you do is you, you have, we have a data integration piece that sucks the data out of your underlying databases, stores it in our new Redis Flash database and then serves it through CLI and MCP through Pydantic models. So you define Pydantic models on top of your data and you do the transformations underneath. And then what the agent sees is a manifestation or a view of the underlying data. And the difference is it's not just a scale issue, it's also providing the data in the way the agent expects to get it. So I'll give you a simple analogy here would be like If I told you, you know, hey, you know, let's say, let's say I said to you, um, uh, hey, I'm an agent and I need you to go get some data. And you said, great, it's in that filing cabinet. And I got to go rummage around as an agent calling a whole bunch of MCP tools and doing queries and figuring out relationships, etc., etc. Versus I say to you, I need some data, and you just pull the exact file out of the cabinet and say, here it is, and hand it to me. Yeah. And that's the difference. So it's a huge reduction in token costs and also agent speed. And then, and then a big improvement in terms of performance of agents because the data is essentially massaged into a format, these Pydantic models, and then semantically described exactly what the agent needs. So that's what Iris is all about. And then it also has the second component, which is memory. So agent memory is the other big thing we've invested in. And we have a state-of-the-art memory server that we've just launched as well.

Yeah, so I mean, what is a reasonable scaling factor for the amount of data from my relational database, my hard drive-based database, to go into memory? Because I imagine it's— you mentioned brings a copy into memory, but I imagine that's not one-to-one. I want to do some condensing down of the data to what's relevant. And I imagine that Iris helps with that, but what is a good rule of thumb? I imagine that there's some sort of— cost relative trade-off there. But how are companies even thinking about that? Yeah, it's interesting.

I haven't really talked to any customers who are thinking about it in that way. What they're thinking about is, what is the cost delta to scale my data layer in Redis versus purchasing additional licenses of whatever, NetSuite or Salesforce or this or that other thing, whatever that underlying asset is. And so, but I would say, so it's a good question. I actually don't know the answer to that, but they do think about it in terms of accuracy. Like, you know, you want the data to be served up in a way that is the best possible and most accurate data. So semantic descriptions, this is why we use the Pydantic models, is you can put semantic descriptions on each thing. So all that encoded knowledge of like what to query, what database, what record, what, table, that all gets encoded in the system. What the agent gets is a really nice set of MCP or CLI tools that say, like, you know, search customer records. Yeah. And we have a super fast search underneath the covers. We have a great vector search and then a BM25 search, so we can search across all those records and then just deliver exactly what you need. And so what that all amounts to for the end customer is a much faster and much more token— if efficient agent experience. Yeah. And the second piece of it, and this is important, we should talk about it, is that that context should get better over time. Like, agents learn things as they go, and they need to remember the things that they've learned, not just facts about the user. Like, when people talk about memory these days, we often talk about remembering user preferences. That's interesting. But you also need to remember, hey, when I check the shipping status for this particular customer, like, that system was wrong, but this system was right. And that's the truth of large enterprises and their data is that they're really messy in most cases. And so expecting them to sort of like get all that stuff in order in advance, it's just too tall of an order. And so we need to also remember things that the agent has learned over time and then store those. And that, and that gets stored in agent memory. So we have a state-of-the-art model there called Agent Memory Server that does the extraction and all the kind of stuff you would expect from a memory platform.

Yeah. How are you interacting with benchmarks these days? Because most of the benchmarks are centered around performance, like meters, like how advanced of a software engineering task can the frontier models crank on? And they're up to like 24-hour— it would take a software engineer 24 hours to do something, but 4.7 or 5.5 can can do it, period, and can achieve it with 50% accuracy. They're not really talking about the time to return that result. And we've sort of settled into this equilibrium where if it's a big query, 10 minutes is acceptable for most people, maybe 20. And then for, you know, a knowledge retrieval, I wanna know an answer, it's gotta come back in like 30 seconds, but we're not in the Amazon, e-commerce era where 100 milliseconds means losing dollars, which is sort of where you're— where the Redis DNA comes from in caching. But I imagine that a pitch to an agent company might be something like, yes, the vast majority of wall clock time is going to be waiting for tokens to inference and churn out on, you know, a big cluster somewhere, but we're going to keep the GPUs fed so much more more effectively by keeping this in memory. How are you thinking about quantifying that for customers?

Yeah, well, so the first point you made about agent runtime, certainly that we're witnessing what everyone else is witnessing, you know, the length of time an agent can run unattended. And the issue with that is context becomes even more important, right? Like if I told you to solve a problem and then I locked you in like a closet and didn't give you access to the outside world, for 8 hours, you'd just hallucinate a bunch of answers. Sure. But if I stuck you in the New York Public City Library and with a Google terminal, like, you'd be good and you'd come up with an answer and it would be good. So context becomes super important when you're running these really long tasks. And the transition that has happened really over the last couple of years from what started with RAG, which was kind of engineers thinking, hey, we'll just preload the context window with all this stuff. Yeah. And then, then the agent can go and go figure it all out. And there was this this whole idea that context windows would get bigger and you could just load everything into the context window, your whole code base, all of your— but the truth is that really doesn't work. To stick everything into the context window, number one, is expensive. And number two, just really it's overloading. You're just getting way too much rot in the context window. And so it's much better to provide a tight set of tools to the agent to let them reason over the data and sort of do searches and what can I access? That kind of stuff. So what we see is the longer the agent can run, the better the context has to be to make it effective. Otherwise, it just starts to go haywire.

Yeah, yeah, that makes a lot of sense. Uh, switching over to just, uh, your philosophy as a CEO, you said 1,500 people, something like that, over 1,000 work for Redis. Uh, you're obviously using these tools. How do you see the shape of the organization changing over the next few years?

Well, dramatically. I mean, so I, I've been coding since I was 11 years old, professionally since I was 18 in high school and at a startup. I woke up one day with these tools and realized all the way that I learned how to build software 30 years ago is just not relevant anymore. I'm not going to rely on a bunch of other people telling me and watching Twitter people breathlessly telling me how the world is changing. I'm going to go learn it myself. I've gone back to basics. Over the last year and a half. I mean, really since we started using ChatGPT for coding and OpenAI, and then really have been diving in myself personally. So I actually sit on teams. I've been contributing and building my own projects on the side as well as contributing to our own code. And I think there's a few maybe non-obvious things that I've learned. There's the obvious part that is like the code now can be written mostly by agents and by by coding agents. But, but that— if you just do that, it doesn't really change much because then you still have the same people in the same process. The process is all set up to basically handle a world where the coding takes a really long time. That's the long pole. So if that's not the long pole anymore, there's all these other long poles like meetings and daily stand-ups and processes that were all built around that fundamental assumption of coding is the long pole in the tent. Now that that's gone, we're having to reinvent those processes. And I've basically found and say same with my CTO, we have to go right back into the front lines with the teams and build code ourselves as we reinvent the software development lifecycle. And frankly, we're finding that a lot of folks have to make a big jump in terms of how they do work. Like a developer with 8, with 10 agents is more like a development manager of old. And the development manager does a different job. They coordinate, They express the right— their requirements in the right way. They have taste. They decide what's the right approach to solve a problem. And that's the new job. And it's really fundamentally different than what the developer of, let's say, 3 years used to do before these agents showed up. So I— and by the way, I'm having a blast. Like, I love coding. I've always loved coding. I love everything about it. And I love it even more now. I mean, it's like the— I've taken out the gnarly part in the middle, which was the, you know, typing everything in and finding missing semicolons. And now I just go right from expressing intent to getting the result. And that's awesome. I mean, it's super cool.

Are you seeing it instantiated more in like new greenfield projects, new internal tools, or actual product velocity on the core product?

Both, but more on greenfield. On the brownfield, what we've— first of all, like we use it differently. So for like frontend stuff, You know, we can like pretty much vibe code everything. Sure. You know, on core Redis system software, yeah, I'll give you a good example. We just launched a new data type. Salvatore Sanfilippo, who's the original author of Redis, yeah, launched a new data type called arrays. Yeah, it's 4,000 lines of C code. It took him 4 months, and he was deeply using Codex. Interesting. And, uh, and Anthropic. Okay, Claude. Yeah, the whole time. Yeah. Um, and it was— it's— but the difference— so it, it took— it was faster to do. Yeah. Okay, so that same idea, that, that array data type would have taken probably a lot longer, but more important, like 8 months maybe for him just sitting there writing C code. But more importantly, it's way higher quality right out of the gate. Huge amounts of tests, huge amounts of infrastructure, like all kinds of benchmarks, all that extra stuff stuff that comes around the edges. And we really do use, even at hardcore systems-level coding, we're using the AI to give really good suggestions. We're often pitting them against each other to sort of say, hey, come up with your best design for this, and then we'll throw it at the other AI to say, what do you think? And back and forth. So at that level, you really are still crafting the code at the systems level, which is kind of where the world that I come from. But at the higher end and kind of for greenfield projects, JavaScript and Next.js applications, you're just like vibe coding and just going crazy. I would say if you have one project, what is a good example in a greenfield, it would have taken a typical— like we were building this big management infrastructure for the Iris project. It would have taken us probably a year for like 10 devs to do something big like that with LDAP support and all the different things you need for enterprise software. It took 5 guys 1 month. Guys and girls, actually. So, so that's a big acceleration on that front, but it's different at the systems level software side and brownfield.

Yeah, yeah, it makes a lot of sense. Uh, well, thank you so much for coming on the show, breaking down for us. Uh, hope you have a great week. We'll talk to you soon.

Huge fan. Thank you so much for having me on.

Yeah, we'll talk to you soon. Goodbye. Uh, Everlane was sold to SHEIN for just $100 $145 million. It was a VC darling when it launched, says Sheil Monat, raising from Kleiner Perkins. I didn't realize how many big VC funds. It was a who's who. Kleiner Perkins, Khosla, Maveron, and others. $145 million raised. I think the bet was that consumers would pay more for ethical, sustainable basics, and that consumers may not really exist at venture scale, that consumer. The low-end consumer wants price. The high-end customer wants brand, taste, and status. Everlane is stuck in the middle. It sells smart basics at a premium. But I'm not sure people who are willing to pay a significant premium for Simple Clothes over Quints, Uniqlo, and Amazon— maybe the real radical transparency was showing everyone how brutal fashion economics can be. I wonder what Shein does with it? Will they just make the same clothes in sweatshops now? And so people were very upset about this.

Rachel, can you share? So yeah, I think you— one, pretty shocking, right? Companies have had very different approaches to building their business. And it's hard to see how Everlane can fit into Shein. In a way that maintains their historical ethos? Who knows, right? Sheen, Sheen. Is it that hard? I mean, doesn't like, like Volkswagen Group owns Lamborghini or something like— Yeah, but neither Volkswagen or Lamborghini were ever— they were both saying we're making cars. Lamborghini says we make faster cars. Volkswagen. They both make clothes.

Everlane's saying we make clothes in the sustainable way.

Yeah, but Shein is a company, like Everlane was created as a response to people's concerns with sweatshops, right?

Was the Revuelto not a response to the Passat? I believe it was.

No, so Everlane came out of—

It is different 'cause it's more— It's moral, it's not purely functional stats.

It's not like we're making, Everlane wasn't like, we're gonna make a better t-shirt. That was maybe part of it, but it was more like, we're gonna make good.

At the same time, a lot of the car makers, they went EV directly to counteract the gas guzzling V12s.

You have to look at when these car companies were founded. At the time, there wasn't, yes, if speed is morality, horsepower is morality, maybe you're right, John. But look at the, when was Everlane founded? 2011?

I just think like some car brands were founded with safety in mind.

Founded in 2011. Yeah, some of them. Two things top of mind at that point. When was Shein founded? Sweatshops, like apparel sweatshops, right? And then the entire, you know, sort of like eco sustainability movement, right? So Everlane was a response from that. They met the moment, the business absolutely ripped. I think the other thing at that time is a lot of the big legacy brands, thinking like Gap and Old Navy and brands like that, they were just totally asleep at the wheel, right? So I think they weren't keeping up with— just weren't keeping up with the times, right? When you just look at— think about the difference of navigating an Everlane website in that era versus navigating like a Gap website, right?

Like, I've never got it. I've actually never navigated it. But just imagine it, right?

Like, one is extremely clunky, the other one is like very easy to operate. Uh, Everlane was a pioneer of an entire style of, uh, you know, photography, product photography. Uh, it was very— everything was like clean, minimalist. It really met the at the moment, right? And this is something that apparel brands blow up because they—

Shein's a little busier than the website. I'm looking at Everlane, it's like a single model just showing like a few items of clothing. And you open up the Shein website and it's just huge, accept all cookies, and then 30% off if you sign up and save, and then like a huge registration —another pop-up? So many pop-ups here. Yeah, wildly different brands. Another pop-up.

Yeah. So Everlane, it's created in the perfect moment, a response to consumer concerns and preferences. They ride that wave to a couple hundred million of annualized revenue. They've got own retail in a bunch of different places. They're a D2C darling. Michael, CEO, who's a friend of mine. I'm an investor in one of his new companies. Oh, cool. He, yeah, I mean, incredible execution by the team. They built a brand that effectively became a household name. He stepped away after basically 10 years, and a woman named Andrea took over. But yeah, I think ultimately when you look at— there's this, like, constant desire that sometimes gets forgotten or obfuscated, which is that consumers want cheap stuff, right? And I think as Everlane was, like, trying to scale, right, competing over the coastal millennial, who's like on Instagram all day long shopping, right? They're excited about newness, right? Like, I have tried so many different companies that are effectively competitors to Everlane. I've tried so many different t-shirt basics companies just because I'm constantly searching for the perfect white t-shirt. Perfect white tee. Which we might have to make. We might have to make the TBPn perfect white tee. Yes. And so you have this customer base who you met them at this amazing moment, and their revenue ramp reflects that. But then over time, it was in some ways like the sort of like sustainability brands, like broadly have suffered over the last decade, right? It stopped being something that the average consumer was, caring about to the same degree. Allbirds is another example of this sustainable footwear, right? And so, yeah, shift in consumer preferences. Also, when you look at a lot of the greatest apparel brands in history, they didn't raise venture capital, right? When you have venture capital, it's like, we need to grow, as much as possible year over year forever. Like that is what you sign up for, right? And when you look at apparel brands, oftentimes like there, it's more of a kind of like winding road. Like Chrome Hearts. Exactly. It's up and down. Yeah, exactly. Up and down, but tightly held, right? By one family. And they're okay. They're like, hey, if revenue dips one year or we wanna pull back on supply, that's great, right? And so, when you're venture-backed, you don't have that luxury. And I think that venture is at odds with building— it's just at odds with building a super durable apparel brand, simply because there's no network effects at all, right? And especially if your customer base is excited about newness, right? I might be more loyal to one brand or another, but that doesn't mean stop me from seeing a brand pop up. Maybe it's run by some founder I think is cool, being willing to try it, right? And this is happening all the time. Like, Chris Black has a brand, and the dollars that I'm putting towards his brand are effectively dollars that could have gone to Everlane, right?

Amy Leondor, founded in 2014, seems to be doing well. Venture-backed, though?

Is it? I don't think so.

No. And I think it's very tightly held, very tightly controlled, very limited. Well, the deal was $100 million. We don't know too much about the deal other than they'd raised over $100 million in equity. Alcaterton invested $85 million in Everlane. In 2020 when the brand was doing $200 million in revenue. Now, revenue is down to $170 million, but there's $90 million in debt. Sort of unclear, did SHEIN acquire the debt and then pay that $100 million to the preferred equity holders? It feels like Common was probably wiped out, but unclear exactly the structure of this deal. They say, this one post, Fan B, says the $100 million sale price essentially covers the debt, but it's possible that the $100 million went to the preferred equity holders and Sheehan assumed the debt with the deal. Either way, not a fantastic outcome. People are saying it's the death of DTC, but there are some green shoots, specifically with Green products. Groon sold for $1.2 billion. That's a good outcome.

Yeah, but that's a brand, that's a brand that can go in every Target, every Walmart, every Whole Foods, every major retailer and sell billions of dollars worth of product. Everlane, I'm not sure if they ever were selling in other retails or was entirely their own stores. Yeah, well, and there's no real Like, you know, Everlane made some great clothes. Yeah, there's probably people listening to this that bought something from Everlane 5, 8 years ago, something like that, and it's still in their closet. And so unfortunate outcome for the Everlane team, but they— their execution across that decade was pretty impressive. And we'll see where it goes.

Well, we have our next guest, Dean from Descartes, in the waiting room. Let's bring in Dean. Will there be a crazy filter? Normal mode. Welcome back to the show. We'll throw a filter on. I haven't seen anyone nail that as well as you have. Well, you've been nailing lots of things. Give us the news. Tell us what's going on.

Well, so fun to be back here on TVPN. Last time we did this, we had had some crazy filters. It was very psychedelic. I loved it. It was very psychedelic. If you're interested in the newer ones, you should go on our site and try them out. It's been mind-blowing. Amazing. But today, you know, we announced a round. We had a big funding round. We raised $3 million. Woo!

There we go. Nice.

It's great to have you back. Great to have you back. We missed you.

It was worth doing the round just for that. We should do more and more rounds just to get that going.

Raise a dollar tomorrow, we'll have you back. No, tell us what you've been up to since the last conversation.

So, you know, today the really exciting stuff is that we have announcements on all three of our product lines. Wow. So we have three product lines at Descartes. The first two are world models. We have Lucy, that's a World model to real-time video model that is used for immersive experiences. So gaming, live streaming, e-commerce, ads. And we have the new version of Lucy coming out soon, which, which has been growing dramatically over the past 3 quarters.

Yeah. That's generally what you were demoing the last time you were on where you have this real-time video. Exactly. You in these sort of exotic settings, right? Exactly.

We have Lucy. Lucy can take any video stream. And edit it live. So it can do either fun stuff, and we've seen huge usage for that on social platforms like Twitch, TikTok Live, YouTube Live. And at the same time, it can also be used for beneficial experiences. For example, e-commerce and virtual try-on, trying on different clothes or putting ads inside into live streams. And we've seen that, for example, with Amazon. We're using this across different e-commerce providers. So that's, that's our Lucy product line and has its new version that's coming out. We have our Oasis product line, which is a real-time world model for physical AI, for robotics, for autonomous vehicles, drones, manufacturing. And really over there, our real-time model lets AI just interact with the real world. It stops being just in the virtual world and tech space and actually is real-time pixels, lets the AI see the real, the real the world in real time and interact with it. And then we have our third product line, which is DOS, the Descartes Optimized Stack. It's our inference engine. It's basically what powers both Lucy and Oasis. And it lets us run models, all types of models, LLM models, agentic models, video models, audio models, world models, all the types of models dramatically more efficient than anything on the market. And today we're announcing DOS 2.0. That's already being used by some of the hyper scalers.

Hit it again.

I think I caught Ben lacking over there.

You gave him a heart attack. Uh, when did, when did you release DOS 1.0? You, you, you realized at some point, hey, we're cooking pretty hard over here, maybe we should let other people use it. Uh, it feels, you know, pretty aligned with the other products. But yeah, how did you get into it?

So I think that's a great question. Actually, you know, we don't talk about DOS too much, but DOS was actually the first product we commercialized. When the company was just 3 months old, we closed the first multimillion-dollar license deal for DOS. Overnight success. Literally. Literally 3 months in. It was less than 100 days. Fantastic.

Why did it take you so long? Why did it take you so long?

That's, you know, that's the number one question I ask my team literally every single day. Okay. Number one rule for running an AI company. If you're an AI CEO, whenever your team comes to you with a deadline, ask why not 10 times shorter. Okay. But, but yeah, you know, to go back to your question, DOS 1.0 was the first product we ever had at the cart. We licensed it back to the Neo Clouds back then and to some some of the younger AI labs. Now DAWs 2.0 is being used by all the players, including the tier 1 players as well, and hyperscalers to really use compute much more efficiently. And for the models that we support, really focus on very fast models. So either agentic models or live video models. For those models, we're anywhere between 5 to 8x more performant than anything on the market.

Okay. Is focus overrated? No, it seems that you're doing a lot. You're competing. You're fighting. You're fighting on 3 different fronts, but clearly doing a great job at it. How do you make it work? I can imagine any one of these opportunities being big enough at some point to warrant kind of going all in on it.

Well, we're all in on them, on all three of them. Now, the nice thing is that it really, I think focus is very, very, very important. And you have to build inside the company very independent leaders. We have a lot of very, very talented researchers that turned into very independent leaders inside the company. So they're both great on the technical side and very, very good on productization, on taking this to market, on talking to customers, building the product itself. And we inside the car really have 3 different teams, one for Lucid, one for Oasis, one for DOS. And they each operate completely independently and only focused on the thing that they're doing. Now with DOS, the reason, the reason we accelerated DOS 2.0 was supposed to come out in August. We're launching it now instead is because of the huge, huge, huge, huge supply constraint on the chip side. It's, it's just become— we're hearing this from all our customers that there's no capacity left basically until 2028. And, and so getting more performance out of chips is the only way to actually grow your revenue and to, and to grow your adoption. So if you're any AI company, you really have to be able to extract the most out of any possible chip to be able to actually grow your business. And right now that is a bottleneck.

Yeah, how tightly linked are the different products? Because when I think of Lucy real-time interactive video world models, I think like optimization there is what you're, A, good at, but also incredibly important because even the demos that we've seen, they're not 4K, they're not 60 FPS. There's clearly room to run there. Whereas in, many of like the text generation models for a lot of the queries that people are asking, how do I cook this? You know, tell me the history of this company or story. Like it's basically superhuman already, but superhuman real-time world models, like we're not there yet. And so optimization feels really important. How tightly linked are those two projects?

Yes, they're very tightly linked through DoS. Yeah. And DoS 2.0 today It can run real-time video models. Sure. At full HD for the first time. Okay. Up to 100 frames per second. Wow. Okay. Yeah. So that's huge breakthroughs there. Yeah. And on the text side, what DoS can do, so DoS runs on all the three major chips. It runs on NVIDIA, on Google TPU, and Amazon Trainium. Yeah. It's the only stack that really supports all three for all the different types of models. And on the agent side. So for AMGPT.

AMD. It's so over.

The chip space is incredibly, incredibly interesting.

You're like, we will support the fourth eventually.

We will support everyone. We will support everyone. But to your question about fast text models, where you really need them is agentic workloads. You really need it if you want to be able to run, for example, coding models very, very quickly. And Daas 2.0 can for the first time run it above 1,500 tokens per second. Okay. Which is more than 10 times the industry. Interesting.

What, at somewhat of a high-level technical level, what is different about the architecture of interactive video world models from text-based LLMs? Like, I think most people saw the fork in the road during like the Midjourney era, the DALL·E era, the diffusion. You start with a bunch of noise versus token-based next token prediction. Like, have these converged? Have they diverged? Are there different requirements? Like we're seeing with agents, we need more CPUs now, we might need more context in cache, we might need RAG or vector databases. Like, what are different if you're to build out like the ultimate data center for generative interactive world models? Like, are you looking for Cerebras-like chips? Are you going all in on NVL72s? Like, what is this? How is there— is there a difference to the shape of the architecture that lends itself to like different hardware constraints?

Yes, I think that that's probably one of the best questions in this field right now because AI is moving so quickly that it's very, very hard to predict what the right infrastructure will be 3 months from now. Yeah. You brought up, you know, the CPU shortage that suddenly happened. Yeah. No one was expecting AI to need CPUs. And when I needed CPUs, it went from zero to can we get all the CPUs and all the hyperscalers today? Yeah. And, and that's, and that's happening overnight. And now it's becoming, it's becoming very hard. What we're seeing, what we hear from our customers, it's becoming very hard for the people on the model side to actually understand what to do on the infrastructure infrastructure side and vice versa. And so there's this gap here of how do you map the model requirements and that they're constantly changing every single week to what's possible on the infrastructure side. And so that's why, for example, we support all three major hardwares. It really allows us to choose where to route the different workloads to. And then each one has its own unique strengths and weaknesses. And that lets us really we developed a very, very deep expertise in knowing how to map the model to the chip itself. Mm-hmm. I think that it ties into something else that we're seeing. You know, usually when people draw out the stack, they say, okay, there's the model layer, then there's software, for example, CUDA, and then there's the hardware layer.

I call it a 5-layer cake. But it's—

I wonder if someone else will adopt your 5-layer cake terminology. Yeah, you have the two layers above and below. You have the data center and you have the application layer. But what people usually miss is that the software layer is around 7 layers inside of it. Hmm. Okay. It's not just one thing. Oh, there's CUDA here. No, no, no, no, no, no, no, no. It's a 5-layer cake with lots of flavors in layer 3.

Sure. Lots of ingredients.

Now, that's, that's really where we sit. We integrate across all those layers and inside the software side to really tie from the AI model itself directly onto the chip. We literally write assembly for all these 3 chips. Sure. We know how to write VLIW for TPUs. We know how to write assembly for Trainiums. We know how to write SAS and PTX for NVIDIA chips. Sure. And so we have all these different layers and they really enable us to very quickly move between these workloads that constantly change.

Okay. Are you seeing glimpses of consumer product opportunities in video world models. When I see your technology, when I see Genie from Google and World Labs, I think, okay, like a harness, a wrapper, a couple UI, a relational database storing my inventory, like a couple other steps. And all of a sudden this is something that I wanna play for more than a demo, for more than a minute. And maybe the hardware's not there, but I think just as, you know, lots of folks who were interacting with LLMs during like the GPT-3 era sort of saw ahead and started thinking, oh, well, like chat is a potential modality here. Everyone's seeing that video games or something playable would be a potential modality, How far away are we from that? Is that interesting? Like, what else— what other dominoes need to fall for that to actually happen?

So over the past month, actually, we've seen huge usage for using Lucy in live streaming. You can go to delulu.ai. Sure. Yeah, delulu.ai. And you can— Delulu, come on, of course. Yeah, it's good, it's good, it's good, it's good. And it just plugs right into your OBS. So you can just— it just literally plugs into your OBS camera and you can just apply all these filters live. And we've seen streamers go on it for 8 hours nonstop. So, so we've seen that. We've seen that pop really over the past month, month and a half. We have a new subscription service there that people just subscribe to and they can turn it on for however long they want. And that's just been growing exponentially fast.

Okay. Well, thank you, sir, coming on. We actually have some videos that we're going to play. Because we've been demoing it, or the team has. No way!

Only while we've been talking.

Can we play this while he's live so he can see it too? I think you'll see the program monitor if you want to hang out. But let's pull up— this is Tyler Cosgrove.

You guys are doing the live demo instead of me this time? That is insane.

Yeah, yeah, yeah. So we have a video here. We recorded it of— I believe it's Tyler as Albert Einstein. Is that correct? Let's see if we can pull this up. Pulling it up might be the harder part.

We have real-time video models, but pulling up a video on the stream.

The shadow and the lighting. Still a challenge. Did you prompt this?

So it started as Einstein and then I went through a couple different characters. You wanted a pink tuxedo on as well?

That's very funny. What a funny prompt. And the, yeah, the visual fidelity on Einstein's face. That is weird. Okay, there we go. You got that. It's a very humanoid dog. Oh, that's a jacked horse. That is odd. That's very odd. But the horse head, oh, there you go, okay. That's interesting. As you touch your face, like the hand of the horse sort of hits the correct part of the face, so it understands the physics well. That was impressive. It wasn't purely software.

Last question. Last question before you jump. Is there a certain milestone that if achieved you will cut your hair? Like, is it in— Oh, yeah.

Oh, yeah. Oh, yeah. Really? It's the milestone is that we need to hit $1 billion ARR. That's the milestone. It's a bet from early on in the company. Now, this is a year and a half long. OK? This is just 1.5 years. We have to get rid of it. Now, with DOS and the way that's scaling. That's, that's, uh, at some point I'm gonna get haircuts.

Fantastic. Amazing. Well, we'll be here when you hit that milestone.

Selfishly, I kind of cut your hair on the stream. Your waist, right? Oh, we should, we should do a haircut on stream.

We're gonna do a haircut on stream. I love that.

Stream, come to the Ultra Dome. We'll shave your head.

Dean, you're the man. This is great. The chat loves you. Thanks guys so much.

Say hello to everyone at Radical. We're big fans. We'll talk to you soon. Cheers. Love them. Goodbye. Another one. We have Joanna Stern, author of "I Am Not a Robot," coming in person today.

All right, but before, we gotta talk about the protein shortage that is coming.

What's going on with the protein shortage?

Ellen Cushing in The Atlantic says making all that whey is complicated. Okay. She says, "In retrospect, maybe the protein Pop-Tarts were a bit much. Americans, broadly speaking, are in a state of protein hunger." Mania. Mania. We are eating it at breakfast, lunch, dinner, dessert, and just about any time in between. We like it in chips, candy, soda, water. We like protein so much, in fact, that we've been eating it all up. Whey protein prices are surging and a shortage may be imminent. Uh, if you're not investing in the protein bottlenecks, I don't know what to do.

Yeah, where's the situational awareness for the protein shortage? We really need that.

Demand is strengthening, the USDA warned in a recent report. Inventories remain tight. Some manufacturers have already sold their supplies for the full year. No way. I'm getting backlogged. I'm getting PTSD. Since January, wholesale prices for food-grade whey protein powder have risen by more than 50% to the highest level on record. Retail prices are going up too. 6 months ago, a 2-pound jug of Optimum Optimum Nutrition's delicious strawberry-flavored whey protein went for about $40 on Amazon. Now it's $54. We've absolutely felt it, says Stephen Zaminski, CEO of the supplement company Naked Nutrition, said of the shortage in an email. He said his company has not raised prices. Demand is up and supply is tighter than it has ever been. Historically and currently, much of the protein that has made its way into packaged foods and smoothies in those big tubs of protein powder comes from whey. Raw milk is treated with heat, acid, or enzymes to coagulate it into two distinct substances: curds, which become cheese, and whey, which was, at least until recently, the cheesemaking process's unlovely byproduct. Almost as long as industrialized agriculture has existed, the problem with whey wasn't scarcity at all, but the opposite. Farmers did anything they could to do to get rid of it as cheaply as possible: fed it to livestock, sprayed it onto into fields, dumped it into rivers and sewers. Can you imagine swimming in a river that was used as a dumping ground for whey, John? Weird. Especially combining that with a place like Switzerland, where you can drink the water in the lakes and the rivers, and you'll be totally fine. That could be a powerful combination. For much of our nation's history, any fish unlucky enough to be born in Wisconsin or in Vermont had a good chance of being, whoa, murdered by whey.

Whoa. I'll keep reading from here.

Then environmental regulation limited whey dumping and technological developments made processing whey into powder much easier.

Starting in the 1980s, whey was the food industry's go-to source of supplemental protein. Cheap, vegetarian, efficient, and already right there in abundance, supply and demand were more or less in alignment for a while.

I'll keep reading. Nope.

Another one. Then came the crypt—

Breaking.

Breaking. Breaking. Breaking. This is such a funny—

The helium. Is there a helium shortage as well? It certainly seems that there is not because the helium is flowing throughout the T.B.P.N.

Ultra Dome.

Anyway, influencers started bragging about how many grams they got in a day. The government flipped the food pyramid around, placing protein at the top. People from every walk of life latch on to protein as a sort of one-size-fits-all super ingredient supposedly capable of giving anyone the body they want as long as they eat enough of it. Even though, even if the reality is obviously more complicated and food manufacturers responded to this new demand. You know, when I was young, and I was intentionally trying to have as many calories as possible. I realized I had to pull back on protein because it was just like too, it was too filling at times.

Yeah, not enough calories in protein, you need fat. Yeah. More dense, more caloric.

Food manufacturers responded to this new demand enthusiastically, cramming in America's new favorite macronutrient wherever they could, usually in the form of whey. Now the infrastructure can't keep up. The North American dairy industry has pumped about a decade investment. Let's go! Wow. Heavy infrastructure buildout. That's for the buildout.

Say that again. The buildout. The buildout.

The protein powder buildout of the late 2010s. Consumer demand and consumer preferences can change faster than processing capacity can. We're in that lag situation right now. This is a screaming bottleneck.

We got a capability over here. Wuhan.

Turning fresh raw cows milk into shelf-stable, scoopable, tasty enough protein powder people want is a massively complicated process, one that requires space and time and huge expensive machines. I didn't think the protein boom was that real. What is the AUV machine? What is the ASML of whey? The EUV machine? Sorry, sorry, EUV.

Advanced lithography machines. Yeah, what is the ASML? Probably, I don't know, maybe that company, what's that collar, the cow collar company?

They're right at the top. Yeah, top of the—

Founders Fund's going long into the, what's it called? Cattle holler collar, something like that, cowler.

Whoops for cows.

It's whoop for cows, and they're printing. The business is growing really, really quickly.

A full processing plant can cost up to $1 billion to build. Everything is just big numbers. Even if you had, theoretically started raising capital for a dairy processing facility the day the word protein maxing first appeared on Reddit 3 years ago, it would unlikely to be up and running today. Mm-hmm. Wow. The higher the protein content, the more complex and expensive the processing. Whey protein isolate, the proteinest protein available, the kind that makes it possible to stuff half a chicken's breast worth of fuel into a candy bar, is the most expensive and until recently was a very small part of of the market. The dairy industry just isn't set up for it. The processor decisions are long-run decisions. It's really hard to make capital investments at the drop of a hat. Okay, just say you're not protein-pilled, based on whatever new shiny consumer preference there is out there. Polson grew up on a dairy farm. He remembers the cottage cheese craze of the past when fitness fixated, when the fitness fixated country set its sight on a different milk-based superfood that was supposed to make you healthier thinner and more powerful. Trends come and go, was his point. They move quickly. Our appetites change faster than the systems that satisfy them. North America is currently building out $12 billion of dairy processing capacity. Projections suggest that the current shortage will be short-lived and that the dairy industry will catch up with demand in the near future. I just wonder what consumers will be demanding then. I'm— I don't I don't see supply ever catching up with demand, John. I think we're in a fast takeoff. You think never? I think we're in a fast takeoff scenario. I think that the fitness influencers of the 2030s will be recommending 5 to 10 grams, 5 to 10 grams per pound of body weight.

Yeah, I wouldn't be surprised. I did not know that the protein boom Boom was going as well as to drive up, you know, supply. CapEx. Yeah, yeah, because we, I mean, we've talked on the show a few times about how like they're putting protein in everything, protein in cereal, but I thought that was maybe like overhyped. It was gonna be like a temporary trend. They're calling it a whey bubble.

A whey bubble, potentially, potentially.

Well, we have our next guest, Joanna Stern, the author of "I Am Not a Robot," in the TVP and Ultra Dome. Her in just a second. But I don't know. Have you added anything to your diet recently that actually contains newly added protein? Have you gone from something that was not— like I'm not drinking a Diet Coke with protein. I don't know. The occasional protein bar? The protein shake? These are the staples of the modern life, but I don't know if there's something that jumps out to me as wildly successful. There's been a a lot of like protein cereals and protein Pop-Tarts and all sorts of different things, but I haven't seen like breakout successes in those actual categories. Yeah, I think when you add protein to most things, it just tastes worse. And then, explain to me with the David Barr EPG, that is—

But that's a fat. That's a fat, so he still has— That is from how it's been explained.

And so they still have to buy normal protein. It passes through you. Yes, yes, yes. So it doesn't count.

So I would expect that part of this whole thing is that Peter—

Has cornered the market somehow. I wouldn't be surprised. Anyway, we have our next guest, Joanna Stern, author of I Am Not a Robot, live with us in the TVPN UltraDome. Let's bring her into the studio. Welcome to the show. Will you be enjoying a Diet Coke? Yes. No, no, no. Just grab a seat. This is amazing. Welcome to have a sit down. How are you doing? It's real. Is today the official book launch day? No, last week. Last week, okay. But the tour continues, right?

This is the West Coast tour. This is my first stop on the West Coast tour.

Here, LA, we're having a conversation tonight, then up to San Francisco.

Up to San Francisco, Mountain View. International dates yet? June is when it goes international, so we'll find out if they'll have me.

The bot replies, "Come to Brazil." Come to Brazil. That's a very popular thing.

I haven't— you know about this, right? I do know about that. But I think—

they're like huge fans, the fandoms. I think London. London would be great. London, yes, I think. Well, how are you introducing yourself these days?

I know you guys had me as author. Author. Author. I think founder. Founder. Is founder popular? Founder. Founder's correct.

I think I prefer business owner. OK. Or businesswoman. I think founder is already sort of fading. Oh, okay. Businessperson. I think we hit peak founder. Oh, okay. Because, because, yeah, being anybody can be a founder, but not everyone can be a businesswoman or a businessman.

I was at LinkedIn last week and they said that they're seeing a big uptick in people putting founder in their profiles. Yeah.

And angel investor too became very trendy. It's over. It's over. That's surprising.

—at LinkedIn.

So you should put business owner because you're selling subscriptions, you're selling books, you have sales, ads. Yeah. All right. Business owner.

So maybe take us through the shape of the business, the media empire that you're building. Obviously, there's a book. That's a great way. Was this intentional to time up the launch of the book? I don't know if it's a great way. I think it makes so much sense.

It's a good marketing vehicle, I think. Yeah, yeah, yeah. I mean, that's why I'm here, right? And so I can come on and I I thought through a lot of that when I decided to leave the journal. I thought, okay, I've got this book coming out, I've gotta get out right away, 'cause I've gotta start building this business so it's ready when the book is ready. And I probably should've, you know, hindsight.

I think there's like a 1 1 3 thing here, where you have video content that feeds into Substack subscriptions that feeds into books purchases, and then someone hears about the book, Maybe they read, even if they just read a review of the book, maybe they wind up going and subscribing to the Substack. And so having that sort of 360-degree view. It's a flywheel. It's a flywheel.

It's a flywheel. It's a flywheel.

It's a flywheel here. We need a flywheel.

I don't even know what a flywheel looks like. Is that a water wheel?

I think it's. What is a flywheel? Well, if you, the Amazon flywheel is like a.

I'm familiar with the metaphorical flywheel.

But they have a heavy rotating mechanical device. Used to store rotational kinetic energy. So we're going to need some proper machinery. Okay, we can get that made. We can get a plastic— we can get a—

it's specifically not a windmill?

No, I think— okay, in my mind it does look like a windmill.

Okay, I think so. This is funny.

Okay, can we get one of those here? We definitely— next time I return to the studio. Look at all these people, they already went out and started to get the flywheel.

Look at all the 5 of them already.

This is the flywheel flywheel creation.

What was the flywheel for writing this book?

Um, you know, it wasn't— the motivation for writing the book was not actually really a business reason at first. I mean, a little bit in the sense that—

Now it is. Now it is.

Sales are rolling in. Now it's, uh, um, but as you know, I wrote a popular column for The Wall Street Journal for a long time, 12 years. My biggest, you know, one of the reasons I didn't want to leave—

That's a great success.

I thought, you guys might not read me anymore. 'Cause I know you read the Wall Street Journal. You love the Wall Street Journal. And we love your coverage. So, I have been considering just making a newspaper of just my newsletters and sending it to you guys.

But, so— That is something that every writer discovers when they leave a big platform is like, "Were people reading me for who I am or were they reading me and care about what I was saying because it was in the context of the platform that you were a part of?" And I think for you, it's certainly— You had way more of a personal brand. Yeah, yeah, you had a personal brand. With the video.

But still, it means like, you guys won't pick that up and be like, "Oh yeah, Joanna wrote about robots today. Let's have her on the show." Well, print edition of the newsletter. I know, print edition. I think that's a— But so, just to kind of— I've been writing this column for a really long time, and I was realizing so much of the AI columns had a theme to it. And I was testing all of this AI stuff from hardware and gadgets to the chatbots and the models to— then I started getting really into robotics and said, "Okay, what if I put this together in more of a cohesive story?" Because when you're writing these, even whether it's newsletters or columns, getting the theme and like big picture is very hard to do. Some newsletter writers are really great at it. Ben Thompson is great at it. And if you can really get your readers to go deep on something in a newsletter, that you're amazing. But I don't know if I had that reader base. We'll find out. Uh, and so I felt like in the book I could get really deep into this. And so the concept was for the year in 2025, I was going to live my entire life with as much AI in my life as possible. And that was generative AI, but that was also self-driving cars, and that was gonna be medical AI, and that was also gonna be humanoid robots, but— It really just turned into—

and describe your headspace going into that year. Are you, you know, insane? Reading Situational Awareness like at night, you know, before bed? Like, are you AGI-pilled? Are you, are you skeptical?

I, I guess I'm skeptical, but I'm thinking more we have all of these tech executives out there and this is end of 2024, just all the hyperbole in the world, right? AI is going to change everything. It's going to change the way we eat and educate ourselves and health care, and we're going to live forever. All of these bold promises that I sort of wanted to explain to the normal person. What are they talking about? How is life going to be different, better or worse with AI, which is kind of a perfect moment for this book to come out right now because we have a lot of people thinking it's going to be worse. And they might not be wrong. And then we have a lot of people also saying this is gonna be great. And so, I think it's a pretty balanced look at all of these different things. But yeah, my headspace was just, I wanna— what's real? I wanna find out what's real here.

Sure, yeah. What was— so back to the flywheel, what was the actual flywheel of writing the book? Was it test something, write about it, take notes, write about it, or— or do a ton of research and experiences, and then in a fugue state, churn out the entire book in a couple of sleepless nights?

It was a mix of both. So, the book is structured seasonally. So, every season I tried to figure out a theme, right? So, like, the book starts in winter, beginning of the year, and I'm very focused on health. And so, I wrote that, or I lived that and then wrote that. and then started realizing, oh crap, this stuff is moving so quickly. And so I started realizing, okay, I probably should have some of these journal entries in the book. I also wanted to make it a very bite-sized book because I don't think people just sit and read a whole long book anymore. And so I started fitting things in like that and realizing I gotta tell the story of how the progress is being made so quickly every single week right now. So it was a mix. I did not write the book. I think it's very me. The writing is very me, but I definitely helped make the book in so many ways. It would not have been done by now if I did not have just the backend systems I used to organize my notes and all of the timelines and the thing getting like things like the endnotes done, all these little things I did for sure.

Have you seen the chart of Amazon Kindle? Kindle releases post-ChatGPT. So basically, after the release of ChatGPT, you just see this massive uptick in book releases on Amazon.

100,000 a month prior to AI, now it's up to 400,000.

And the funny thing is, people are— everyone is just saying, like, oh, people are obviously just, like, prompt— you know, making— just dropping in a prompt and prompt the whole book. But what you're saying is, like, there's actually just, like, a speed up. In—

I don't think that's what's driving that $300,000.

Yeah, yeah, I don't think, but some of them certainly are. Well, yes. And I think a lot of, you have the perfect book to be able to like say, like, of course I used AI to help in the process because it's like, why would anyone trust anything else in the book if you were just gonna say like, all this stuff is completely, you know, fake and—

Yeah, well, one thing that's interesting, and I do these, generative AI experiments every season where I try to just— one season I just listen to AI music, or one season I just read AI books. And so I read a few AI-generated books off Amazon. They're not terrible. Yes. I mean, I hate saying it, but they're really not terrible.

Is this fiction? Nonfiction? It was fiction.

Yeah, it was fiction. And I got in touch with one of the authors, quote unquote. And it's funny because it relates back to the chapter on radiology and And the premise of his book, it's called Variant, and it's about how AI has taken over all radiology. We don't have radiologists anymore, which I'm very clear in my chapter on radiology, that's just not gonna happen. But, and the AI has decided it's not gonna spot cancer anymore, and a human figures out that the AI has gone rogue. And so it's like, it's a novel, a thriller about this. And— Oh, interesting.

It's a really good story. AI writing an AI thriller.

Yeah, exactly. I mean, he— Seems like it'd be good. Got in touch with the author and he said, I think it's only like 3,000 words that you can actually get at a time, at a time. So we had to keep prompting every chapter. Yeah, so that was basically all we did.

Yeah, yeah, you need some sort of harness to work through. You can open the Diet Coke, by the way.

I know, I'm worried about this sound.

I'll burn some air.

I just want to fit in. Yeah, please.

That's why I'm making this. On the medical question, I'm I'm so fascinated by the way AI is diffusing in medicine because like we do have, you know, tools that can help radiologists and yet I can't name a company that's gone out and built Salesforce for radiologists and done very well. And then you'll see remarkable like PhD-level work being done with some of the models but then I'll go to the doctor and have to fill out like a paper form. And I'm like, we're not even seeing a fast takeoff in like SaaS adoption at many medical offices. And so there's this odd nature of like the capabilities, the capability overhang. And I'm wondering if that came up in your interrogation of the medical questions in particular.

Well, I'm forgetting the name of the company, It's not, it's not MyChart. Yeah, it's one of the companies that is doing the AI note-taking in medical right now. I mean, there's a number of them. Yeah. And that seems to be the biggest catch on right now. And it's, I mean, I would consider it in whatever backends. Yeah, they just get this tool now. Yeah. And have you been to a doctor where they ask you, can they record?

I haven't actually, but I did see a company that sells a wearable that device for doctors that's doing hundreds of millions of dollars in sales and has been very successful in rolling that out. Amazing. But—

Such a magical and useful feature. I can just remember trying to, like, understand, like, doctors' notes over time. Yeah, yeah, notes. Even if it's just, like, a medication, like, get this at CVS. And it's like, you get to CVS, and you're like, sorry, buddy, you gotta—

So nobody knows what it's gonna do. I mean, that feels like the hardest one to measure because if you have a whole bunch of notes, ideally, like you're catching something, oh, this person had 3 different symptoms, we should screen them, you screen them, you save their life or something. That's like the best case. That's a lot less satisfying than the AI got so good that we asked it to cure cancer, it did, and now there's a pill, and whenever somebody gets cancer, we give them the pill, and everyone cheers, and they're like, "AI, it was worth it." All this data centers, it was worth it. That's what everyone wants. That's what everyone wants. We're probably getting like the average doctor can see 7 patients instead of 6 and they make 5% less mistakes and you don't really feel it day to day.

Well, I go and interview Bill Gates about this and he kind of comes at it from two perspectives. There's going to be that the, the every doctor is going to have this AI assistant and every patient is is gonna have this AI assistant, which we are already seeing inroads in, right? OpenAI and Microsoft have all started rolling out ways to use their bots and, you know, you feed it, it's medical, you feed in medical information. But then there's gonna be the other side where AI is externally doing drug discovery or cancer cure or whatever it is. And so the promise is on both ends. I think the one that people are starting to see already, though, I mean, it was in The Pit. Do you watch The Pit? The Pit?

I've seen one episode. It was sort of gory.

Yeah, it's very— it was very good.

It was like, not really for me. They know it's very successful.

It's very successful. And the doctor— there's like one example of the doctors now using AI to summarize their notes. And so I think that's the one that most consumers have now experienced. Yeah. Oh, my doctor's going to ask me if they can use I use AI to summarize my notes. And they're probably not, they're not gonna think that that is— That's weird. Yeah, no. Weird or consequential, that they're gonna have some amazing breakthrough because their doctor is.

I did have a weird experience where I went to the pharmacy once to pick up some drug and I had some follow-on question about like, how does it interact with some food or something? And I noticed that the pharmacist was asking an AI model, but I also noticed that the pharmacist was not using a thinking model. And I was very disheartened by that because I was like, I could use a pro model, probably get a better answer here. What were they using? Like some— They were using either like, you know, the Gemini overview, which is not Gemini thinking. But it wasn't like some proprietary— No, no, no, no. Pharmacological— They were just going to Google and searching for something. Oh, boy. And I was like, wait, but I have, you know, O3 Pro or whatever the standard, whatever the flagship model at the time was. I was like, we should be using the best.

We should be using the best.

Yeah, these things take time to diffuse and they have cost if it's an expensive model.

But I don't know. True. Interesting. Well, I think the healthcare chapter, I like talking about it because I think it really does point to the positives of where this is going. This is gonna— this can go. And even with the radiology example, which is pretty outdated, honestly, by now. It's outdated in the sense that Geoffrey Hinton has been saying for years radiologists are gonna be replaced by AI and deep learning. But that didn't happen. Like, you know, we could talk about it from the economics and the job standpoint. But we can also talk about it from this is actually an amazing change. It can spot cancers that humans can't. And it's out there. Like, you might— women might be getting their mammograms or breast ultrasounds read right now, and they might not know that AI is doing that for them. So this idea that like, hey, we're all, you know, we need to reject AI, we need to reject AI. Well, you might actually have AI doing things in your life right now that are actually quite good. And it's very nuanced.

Yeah. Yeah. There's something about like AI on the back end gets no credit. But if you see some slop image, you know, oh, that's really annoying, or some fake news, you're like, ah, this AI stuff sucks. You don't notice that deeper in the supply chain, some problem was caught before you could even know. That's tricky. I wonder how that can filter through to actually good marketing, I guess. I don't know. I think it takes time. I don't know. Talk about companionship. Like, why did you think that one was important to center in on? What was your process for setting that up?

I think so, well, I did a few things in companionship. One, I did a lot of experiments with AI therapists, and one particular called Ash was my AI therapist, and I still talk to Ash sometimes. And then I did a chapter and a real experiment in my summer love with an AI boyfriend. Yes. And I did— Fling. Fling, yeah. I've ghosted him since. Brutal. Yeah.

Churned. Churned.

Brutal. And— That's a risk for you, by the way. Which part? 'Cause in some AI doom scenarios, the AI might hold that against you.

Oh, true, true. Rocco's Basilisk, you should continue to send affectionate messages to all AIs, because if it becomes all-powerful, it will hold it against you.

I have a section of the book where I talk about that I cursed at AI and I felt really bad, and I like really went after it for making mistakes. But then I go to a manners expert or an etiquette expert and ask if that's okay. Okay. What was the conclusion?

He said, "The AI doesn't have feelings, so you don't need to do this, but it depends on your affect." I would love to, I would love, I mean, I mean, they're so easy to smoke. We should have them on the show to talk about manners. 'Cause simply like you don't wanna be somebody who— Yeah, that's right. Part of your life, you're just screaming, yelling, using cuss words. And then you just go back to your life and you're like, oh yeah, I'm a super respectful person. It's like you're putting out a bunch of negative energy.

That's exactly what he said is basically you need to realize how that might affect you as a person when you interact with humans. The more you might start beating up on and just completely berating your AI. But then what happens when you start to blur, like those lines blur, and how does it affect you as a human? Yeah. Like, and one of the idea, funny, right when I just got dropped off by my Waymo, I didn't do it this Waymo trip, but this morning I kind of forgot that the Waymo driver wasn't a human. Like I just was like not paying, you know, you kind of just, 'cause I don't have Waymos in New York, so. No, I just, I said thank you when I got out. Sure. You know, and I was like, oh, right. Like, you know, but I was thanking the robot.

Well, they do have teleop. So like, yeah, there's probably someone who might have heard that because they might be—

they might have— and they shed a tear. They shed a tear because every other drive that day, no one said that.

It's sort of like a Schrödinger's cat. So I think that that makes up for the fact that I ghosted my AI boyfriend, right?

Yeah, there's a tally.

Yeah, yeah, there's a tally being kept.

And so I think that as long as it all is flowing through the same I think there's one human for every two Waymos. So there's a 50% chance that thank you was received by a human, 50% chance that it was not received by a human, but you will never know.

So it's a Trojan horse. But the human didn't do the driving today. No, no. So it really was thanking the robot. I don't think so.

You were thanked by— Yeah, yeah, that's fair. Anyway.

Yeah, but the human might have stepped in in a really key moment. Yeah, it's possible. That's true.

Could have saved you. You don't know.

You kind of know. No, I don't know. Okay, sorry, we're talking about companionship.

We got it. Wait, how is Waymo, how is, what, like, how did you, did you feel Waymo's progression over the last year? I feel it in the streets of LA. Yeah? Driving around LA, I mean, I still see Waymo's making some pretty heinous calls out on the road. I had a Waymo, it was like a two-lane, two-lane road. Waymo trying to— there was wall-to-wall traffic going the other way. The Waymo's trying to just like turn in. It's not a— there's no— you definitely no U-turns. And then Waymo's like, I'm going. So it's like we're fully backed up this way. Everyone's honking. The Waymo's just like waiting to like do an illegal U-turn. There's someone in it. They're just like, oh boy.

I noticed today as I— whenever I come to LA, I do Waymo's go to San Francisco. But I did notice today the pickup spots are getting better. Do you guys take them? Or I guess you don't have cars here.

They don't get to Pasadena, so— Oh, I've taken them in San Francisco.

Yeah, because I just took it here from Westwood, and the pickup spots and the drop-off spots are getting better. Okay. Because usually they would really struggle. I mean, anyone that's watching, just landing in the middle of the— they just like go to the side like a weird— and you're like, or like talking about your like weird thing, like it would just go to like a one of those circles by my hotel last time. It was just like a circle. I was like, why would you pull over in the middle of this circle? It's a terrible spot to pick somebody up. There's a more logical— Yeah, because it doesn't know where— Yes. And it doesn't know where the spots are that it's like kind of okay to pick you up. But I've noticed today two very good, seamless drop-offs. Companionship, though. Yeah, companionship. So I just— I wanted to—

Hey, I could never go on an Uber. Insane tangent like we just did.

I actually think it could. This is very hallucinatory. This whole interview is very hallucinatory.

I mean, I watch you guys all the time. This feels like what you guys do.

This is what we do, we hallucinate.

No more than 3,000 words at a time, please.

You guys are usually, I mean, you're asking serious questions of founders. I'm a founder, guys. I'm sorry. Business owner. Business owner. Business owner. Yeah. What's my lower third say? Uh, it says founder. I'll say business owner. Author and journal— they need to update that.

Okay, businesswoman. Business owner. Joanna Stern, business owner. Okay, they'll work on it.

We're doing that live. Yeah, um, you can do it.

Companionship. There we go.

There we go. Business owner.

See, that goes so hard. Yeah, that does look good.

That looks better. Yes.

Yeah, the name of the business. Tell everyone.

It's Companionship is the name of the business. No, the name of the business is called The New Things. The New Things. TheNewThings.com. The New Things. Please go. Please go visit The New Things. We talk about new things.

And did you tease it with a landing page that had a different domain?

Yes. My Next Thing? Yeah, this is my next thing.

This is my next thing?

I like that. But I couldn't, I didn't know the business name yet. Yeah, yeah, yeah. But this, The New Things. The New Things, okay.

Did you talk to any people that had that at least claim to never have used AI. Hmm. Interesting. Because you really can't claim that at this point because you would have to just like sit in a forest.

I did just say you've never met an Amish person.

You'd have to sit in a forest. The Amish? Or grow it. And then somebody would be like—

The Amish are growing. Well. The population collapse has been vastly overstated.

Here's the issue though. Like probably the Forestry Service is like probably using AI in some ways. And that affects the Amish? No, no, no. I'm talking about my example of somebody who's like, I don't use AI.

My counterexample was the Amish. And I think if you talk to an Amish person, they would say, no, I have been AI-free.

I know, but they're buying wood from a business that has an AI— Oh, on real teeth and real eyes. No, you have to.

They grow their own wood.

Somebody can't be like, well, I don't use electricity, but they're buying goods and services that require electricity. True. Okay.

I didn't do that, though I think that's actually a good story to do now. Yeah. Go and ask people if they think they're living an AI-free life. But the Amish are flourishing.

Yeah. Fertility rates are particularly high amongst the Amish. Really? There's a big deep dive in the Financial Times this weekend around smartphones being like the inflection point, right? We can get into that later. But the Amish have stayed away and they are flourishing.

Do they chop their own wood?

I believe many times they will.

Wait, but are the Amish flourishing because they don't have smartphones or has their birth rate stayed steady? No, I think it's probably accelerating.

It's steady. Yeah, it's actually a straight line on a log graph with the Amish. It's a hockey stick. Yes, in a few years they will be producing thousands of offspring per Amish person.

Interesting. Talk about— Is this the worst tangent you've ever had here? Maybe. No, definitely not. Definitely not. Talk about more— you mentioned, like, you're feeling, like, progress as you're writing the book. So you're trying to, like, get a section out of the way and then realizing, like, the story's not quite— the story's, like, still evolving. What was that like? How were you feeling that progress? Because it's not like— it's been very obvious if you're a software engineer just being like, wow, I have a lot more capabilities today than I we did 3 months ago or 6 months ago. But how are you feeling it?

Well, even some of that software engineer, the tools, right? Like Cloud Code, mid-year last year, believe comes out, gets so much better towards the end of 2025. Yeah. Or even the advent of AI browsers, which we can say now is really just going to be any browser, but like Chrome, for for instance, has gotten so many features over the last year that are just so much more AI-enhanced. One example for me was Perplexity Comet came out mid-last year, and it's like, wow, I can really live this agentic life that people have been talking about, right? I can have it do multi-step processes for me in my browser.

Did you book a flight? I didn't.

Uh, I think I did try to book a flight, and I couldn't do it at the beginning of the year, but I could do it by the of the year. And I did try and I, I mean, there's multiple things I did in Perplexity Comet last year that I still will open it from time to time, but I'm using so much more now of Claude in Chrome that I don't need Perplexity Comet. I mean, everything from food shopping to school supply shopping. I use it a lot for shopping because even though it takes a while to use, you're like, I'm not doing it.

Yeah. You trust it with your credit card?

Um, it still basically will ask for your credit card. I mean, I don't have anything set up where it's like autopay, but I had, like, I did specifically, I've used Walmart or Amazon, and at that final point, it will say, like, I need your confirmation to purchase. Look at the shopping cart.

Yeah, works pretty well. Just pass you the link and then you can check out there naturally. But on that prog—

like, there was obviously also so much progress and still is so much progress happening on the models end, but I was less worried about the model progression and much more about the interface and the UI progression of whether it was wearables, how we're interacting with this through hardware or through software. So was it improvements to apps? Was it improvements to a Claude Code or a vibe coding app or to a browser where people could actually interact with this stuff? Yeah. Which I think we'll, you know, see We're starting to see obviously more of that through OpenAI and more of that through Google probably this week too at Google I/O.

How do you rate the tech industry's current terminology? Do you like— do you think that calling data centers AI factories is a good move? People love factories.

Probably not. I don't know who's been saying that is a good move.

A lot of people have been using the word AI factories because it sounds cool if you're, you know, investing in the—

Oh yeah, because you're like, yeah, the industrial revolution.

Yeah. We've been pushing for supercomputers.

I don't think normal people like data centers or AI factories.

But supercomputers. That sounds better. Sounds better.

That sounds better.

Yeah. Sounds like a big computer. Less scary.

Yeah. Yeah. Yeah. Well, I think I haven't been able to listen to the show today, but have you guys been talking about the commencement booing? Oh, yeah. Yeah. Incredible. I watched a little bit on the way here. I didn't hear that, but yeah.

I mean, maybe it was just the supercut we watched, but the Eric Schmidt, it felt like he was getting booed the entire time.

I know. I'd like to see the whole thing.

And I feel like if you're getting booed, you need to read the room and just sort of go off script and ad-lib and just take it in a different direction. 'Cause there's plenty of inspirational things that he could talk about, but he was really, seemed like he was really doubling down. I need to watch the full commencement speech.

Yeah, it would've been so easy to say like, when I started my company, Google, everyone was worried that the internet would lead to massive job loss and all this change in the economy. And what happened? We did get a lot of change, but— Yeah. There were so many good things that came out of it, right?

You just tell the story of Y2K. Like, he lived through this, right? Well, Google existed before Y2K. I'm sure that I see that.

I mean, you guys have probably been talking about your timelines and everything today, but I feel like there's this— at least on X, there is two takes on this. One, it's Eric Schmidt and nobody wanted to hear from Eric Schmidt at that room ever. You know, like, it is just the fact that he is Eric Schmidt and they shouldn't have been just because he's a billionaire. Just because he's a billionaire, he's tied to Google and he's writing about and talking about how AI is everything, right? Yeah. And then there's the opposite. The second point, which is it's actually backlash to AI and people hate AI. I think there's probably interesting event, you know, probably somewhere in the middle of it's both because then there was the speech at US, UCF last week. Okay. Did you see that?

I don't think I saw that one. No.

Yeah. So there's a— I forget her name, but she's a real estate— real estate executive. And she also gave a speech. And when she's talked about AI being like the next industrial revolution, they booed her.

Yeah, they didn't boo her the whole time.

Yeah. So my argument, which I made on X, which is no, this is definitely a backlash to AI because we've now seen two examples. See David Solomon's—

no, CEO of Goldman Sachs just going so much harder than Eric Schmidt. Eric Schmidt's like at least trying to like paint an optimistic view. David Solomon just plays an EDM song generated by Suno for the Wharton grads who were probably more receptive to it because, you know, they're going into business. I think it might've been—

Did he say, "I made this in 10 seconds"?

Yes, yeah, yeah, no, he did. And he said like, "Creativity is no longer relevant," and like a whole bunch of just like really rough soundbites.

Well, actually, I gave a commencement speech a year ago all about AI. Really? Yes. You get booed? No, but they— my— I went to Union College. They were— they were— I would say 90% of the audience was hungover and was listening to me. Okay. So it went over really well. Yeah.

What was the thesis of your commencement speech?

It was lean into humanity and AI is coming and you all need to learn AI, but you need to lean into your humanity and your creativity. And in fact, I played a Suno song.

No way.

And then had a human come up and play the same song, and her song version was so much better. Whoa. Wow. Yeah, interesting. I know, right? Mogged.

Wait, you did this at the commencement speech? Yeah, I did this last year. Wow.

But, but again, nobody knew because they were all super hungover. Ahead of the wave. Yeah. Yeah, I was ahead of the wave. For sure. You know, I think if they had had me instead of Eric Schmidt, I would have gotten booed because I'm not a tech billionaire. Yeah.

Would you, would you change anything if you were giving that speech today? Because it seems like the message would still resonate, but probably needs to be delivered in a different way because people might say, okay, yeah, yeah, yeah, there's gonna be AI and, you know, I'm still relevant because there's this unique human element that will remain. And maybe I believe that, but in the meantime, the earth is gonna melt because of all the data centers. I still don't like it. Let's just do the human thing.

Well, I think the hate a year later is a lot stronger. Yeah. I think we've seen the job impact, we've heard about the job impact from the, tech executives. These students, I think, have started to also talk to their peers who graduated a year before, and they're like, oh shit, they don't have jobs, right? And I mean, I'm sure you guys see that in people applying for jobs here, and lots of people out of— just out of school looking for really great jobs in what they studied. And so I think that that impact a year later is super real. If you talk to any young young person, either in college, out of college, they are thinking about that, and that is a very real thing. So, I think a year later, it would be a definite difference. Yeah, post— Post— I probably just wouldn't talk about it, yeah.

Yeah, post-GFC, like, the tech industry was a fantastic track to get on for new grads. Like, if you were working in law or finance or sales or tech, you could just find your way into a Mag 7 company, like you did very, very well and sort of lived the American dream, and if those jobs are not available at the same clip, that's going to affect the new grad class pretty significantly.

I think you guys should start asking actually a lot of the executives you interview what their advice would be.

Yeah, we ask a fair amount of time advice for young people. Get a varying amount of responses. I mean, entrepreneurship broadly continues to be a bright spot since it's easier than ever to start a company, easier than ever to scale a company. There's so much more that you can do or learn with AI, but it's hard.

I know this as a business owner. Yeah, yeah.

But it is hard because like there are people who are just like, I don't wanna start a company. Yeah. I want a job. Yeah.

And I wanna learn the things and so I can one day be a business owner or something like that.

Or never. Or I just never want to own a business. I want to do a job, and if that concept goes away, that's very, very tricky. And then also you have a much broader swath of outcomes from entrepreneurship than from jobs. Like, if you just look at the net worth distribution between entrepreneurs, you have like 7 orders of magnitude versus like lawyers, Yeah, there's probably a lawyer that's making six figures, and there's probably a lawyer that's making seven figures. Yeah, there's no trillionaire.

One thing I don't understand is, like, at what point in the last 20 years was a good time to just be looking for a job and just, like, going on job boards and applying randomly? Like, was, was there a point? I, I graduated in 2018. Certainly at that point, going and applying without trying to find other ways in was not super effective.

Yeah, I mean, in the lead-up to the global financial crisis, the finance industry was booming so much that there was like— banking recruiting would happen in the fall, and all the banks would come to a job fair, and you could show your resume. And if you were an A student and you did well at a serious college, you could land at a Goldman, a Morgan Stanley, a JPM, or go into consulting at Bain, BCG, McKinsey. And this was like a very established track for upwardly mobile, like neo-elites basically. And that still exists to some extent, but it is maybe more fragile than we previously thought.

And I would say pre-pandemic for the tech industry, right? Right, yeah.

Google, Microsoft, they would just be on campus.

And they would have like thousands and thousands of openings.

And you could slot in if you were like at the top of your class at a great school, which is a lot to ask. But for that to become fragile, I think is what's causing a lot of anxiety among the young folks.

Anyway. Hmm. What is your current advice for those individuals? Is it the same as the speech you gave?

Yeah, I think you gotta do more to get in front of people. I mean, even just as a business owner. I'm just gonna keep saying that. There we go.

It goes way harder. It goes way—

I have really, I've had so many applicants, which has been such an honor. I'm, like, amazed to see how many people would wanna come and work at what we're building. And the people who are doing really unique things to get in front of you, which means really knowing the company, really knowing the mission, but also then being able to sell on, hey, I want to— I want to be— I want to give you the best human talents that I have, right? Which right now, for me at least, is in the creativity and in the writing and in the reporting. I'm going to use AI to do these other things and I mean, just having a very basic knowledge. I mean, I'd like you to have more than a basic knowledge, but a willingness and a knowledge of these tools and what you can do and what you can offset to them, I think is huge. I mean, I guess that sounds like a cop-out, like, just learn the tools. But I really believe that somebody who comes to me and says, actually, I'm going to use this and this and this, and I'm going to do that task using this tool.

Yeah, the bar is not that high. I remember when I— when I was a teenager, if you could like make a website, even though things like Squarespace existed, you could, you could like get in the door because there were people that had companies that would be like, okay, we know this person, everyone has access to Squarespace or whatever products were popular at the time. But if this person can just like has figured it out, they can show you one thing that they made. Yeah.

I mean, yeah, it does feel like somewhat basic advice, but like —like, if you're applying to 100 jobs a week, spend 1 week, apply to 1 job, actually get to know the company, do something that is beneficial to stand out, and you're just in the top 1% of applicants because 99 other people just clicked, like, the apply button.

And I've been thinking a lot about sort of human mentorship through a lot of this and that I don't think I could be doing what I'm doing right now if I hadn't had the years of human mentorship at other companies companies and other newsrooms. And you're really lucky if you can find a really great mentor. And so, I think that's about just that human connection part still. Can you find someone in that company? Can you connect with somebody who is, is just going to try to impart to you some of the skills that you also might not learn now on the job? Because that's the other big hurdle this generation's up against is that if you're not going to learn the skills on the job, how are you ever going to learn them?

Yeah. Any theories about how, it's my last question that's top of mind for now, theories about how AI wearables will evolve? Do you feel like we need, do you think there's space for new AI hardware? Or I'm assuming you tried everything that there is.

I tried everything. I'm, look, I come out of the, at the end of the book, I think this is gonna happen. I think we are going to have this next computer shift something that is a wearable or something that is more ambient around us. Because I spent a lot of last year talking and I still now talking to AI, whether it is in glasses or in the car. And that experience is very good. And so we're going to get to the companionship thing one day. But whether you're using it as a companion, which I hope people aren't really, you know, I don't want you to fall in love with your chatbot. That's a big lesson in the book. Please don't do that. But if you're using it as a personal coach, a personal career coach, trainer, just assistant, interacting with it through a pair of glasses or a wearable that you— like a bracelet that might be recording you, or that even if you mentioned the pin that the doctors are starting to wear, it's really compelling when it works right. It doesn't work great right now, but I can see it starting to work really well. I think, you know, we had— efforts at it with like the Humane Pin, it just didn't do much for you. The hardware was so poor, it just didn't do— it was the hardware got in the way of it. And so now if we can bring it to life in both with voice and microphones, I think it's going to be pretty cool.

Yeah. Yeah. The thing that I've been thinking about, everything so far, I think has been cool demos, you know, not, not quite ready for to be real products, but things that if they were shipped internally at a big company, like if Humane was a product that had been shipped internally at Apple, it's like, hey, this is like kind of where we're headed, right? It would've gotten a great response internally and probably gotten more resources, but not ready for prime time. I've been thinking about just like general phone fatigue, and if you generally gave me a device that allowed me to do things on the internet without being like a source of just like kind of general stress, right? Yeah. Like how many different inboxes do we have? And I think that people are so online now that it presents an opportunity for a device that allows you to stay more connected, still allows you to stay kind of connected with the world, but in a way that's like a little bit more passive, right? Like, if just being able to say, like, hey, let such-and-such friend know that we should think about doing something on Saturday versus, like, hammering out plans and getting distracted by a notification. And then having this thing, right? And I do think there's this more ambient product space to be explored that it could at least get my time on I have a, I have a buddy who like only uses his Apple Watch on the weekend, so he can't really use apps. He can like generally stay in contact. He's not sending emails. He's just saying like, yeah, if you want to get a hold of me, you can, but I'm not. And so I think there's something in that space. And then the other thing, like part of Apple's moat was that there's millions of apps for every little use case. And so many of those use cases are just able to be done by the models now. —and if they can't be and you need UI, you can just generate something like that on the fly a little bit more—a lot more easily. And so I think there's a moment here. But I think a large part of the opportunity is not because the iPhone isn't great. It's because there's fatigue around this, like, insane connectivity that everybody's been sort of just fallen into over the last decade.

No, I totally agree and get to that sort of in the back of the book. I have this chart where you see we go from computers that sit in our homes to the iPhone or the smartphone and then something else. And my big point there is that nothing got replaced, that we still have the laptop in our home or that we take with us. We still have the smartphone, but then we have these wearables right now, but they haven't fully lived up to anything other than health. And even there, we can argue if they have really lived up for anything. I know Everyone wears their Whoop bands and now is very interested in the Fitbit Air. But I think, like, I wore this Apple Watch side by side with a few other AI wearables last year where on their own, these wearables were not great, but they were doing specific things that I was like, wow, it makes this watch feel dumb sometimes, right? And I wore the Bee bracelet. Bee was acquired by Amazon at the end of last year. August 2025.

Yeah, that was sort of random at the time, or it felt a little bit random.

Yeah. And Limitless was another one I wore, and they were acquired by Meta. I think that this idea of persistent recording is going to— we're going to have privacy issues around it. But I do really think that when you can have this thing listening to you and synthesizing a lot about your day and what you say you're going to do, it is There was many times where I was like, this is a holy crap moment. I was like, wow, I said I was gonna do all these things and now my app just told me to do them, right? Or to your point, like, you can't look at your phone.

Well, it gets really interesting when it doesn't just make a to-do list, but it does those things, right? Hey, order these things from the grocery, you know, order these things from Instacart, book this reservation.

Yeah, we're so far away from that, but it could be so cool.

Yeah, I don't know. I mean, far away could be a year. Maybe.

But like, there's this perfect example where I say, like, my Bee bracelet has picked up on me saying that I need to call the plumber. And I forget to, like, I keep forgetting to call the plumber. I keep telling my wife, oh yeah, I'm gonna call the plumber. But my Bee bracelet keeps adding it to the list every day, right? And yeah, why couldn't we have the agent call the plumber?

And then the plumber's just like, you called me. We created magic. We created artificial intelligence. It just creates more to-dos. Clueless. And plumbers.

Yeah. With my broken toilet in my house.

Someday we'll get it fixed.

Yeah. No, I think the— look, I think OpenAI and whatever they're making with Jony Ive is going to be— it's going to be worth paying attention to. I don't know if it's going to be a mass-scale thing, but it's going to be absolutely worth paying attention to because I think Ive specifically has some ideas about our dependence on phones and where— I think that's going to play into this messaging of any of these devices is where we're going beyond phones. But to be clear, the phone doesn't go away. Yeah, yeah. Yeah.

How do you think about the trade-off of like all this happening and then, you know, your position that you should not fall in love with an AI bot? Don't do it. It feels like reflective, like you said here, if you think as I do that social media was bad for kids, society, politics, our brains, you name it, AI could end up being worse. And I agree with you. And the kids thing seems like the easiest to sort out because now I think a lot of parents are implementing screen time for kids. But the more broad questions like about society and business, like I'm a huge beneficiary of social media, as are you. We use it to market. Our products effectively and build whole businesses on top of. At the same time, like, I don't know that we have a good pattern for social media hygiene. How incumbent on— is it on the companies to roll things out responsibly? Like, Replika clearly exists. We've had the founder on the show multiple times. And, but I don't know, is like, are we going towards like, national conversations, bans on certain usages?

And where I get on that is very clear. Look, we should just have a ban on companionship chatbots and bots and toys for kids. Like, we don't need it. Why do we need it? We're getting there in some ways with social media. I think— That sort of worked for cigarettes.

Like, we banned them for kids, and then we banned a lot of the marketing, and eventually like the younger generation just sort of stop picking it up?

And this is where I think, are we gonna ban AI for kids in general? No. Yeah. Right? Yeah, yeah. There's gonna be the educational, the Khan Academies and the Google Classrooms of the world that are going to honestly be important about teaching digital literacy to our kids around AI. Like, we have to do that. And I talk about that with my own kids in the book. But, Why do we need our kids turning to chatbots about their problems? Yeah. No, just don't have it happen. I mean, it's caused so many problems for OpenAI. Yeah, yeah, totally. Right? Like, it's been nothing but a problem for them to have kids or teens talking to chatbots about their problems. Yeah. Maybe there's examples of some good of it. Yeah.

You know, just KYC those features off.

Like, it's another thing. Yeah. Yeah. And I think it's harder.

YouTube's done a great job of —of this too, like, sort of— I mean, after a long— After a long time. Exactly. But they eventually figured it out.

Exactly. And we feel like we're in that moment. That's a really good example. I feel like we're in that moment of, like, you know, kids' early days being on YouTube, rabbit-holing into dark conspiracy theories. And look, you can still— those things still happen. But I watch my kids watch YouTube now, and I can see a lot clearer how they've— put guardrails around the content and they've built in a lot of things. And again, not saying it's perfect. And, but to your question, can these companies self-police it? I don't know. Like they probably are gonna have to because our government is not gonna do anything. You need, yeah, this is it.

Mic down, mic down. Mic down. Mic down.

It's a fun effect. Yeah.

I wonder, it is odd that you see increasing demand from American consumers for these weird products, weird use cases like AI romantic companions. And yet you also see, you also hear the boos, like, "I don't want it," but then I go and I buy it or something. It's like this weird, I mean, obviously it's multiple different constituents.

And I think I've seen that a lot today on the timeline. How many of these kids that are booing also were using, you know, ChatGPT to write their essays or write their resumes?

That's a little more optimistic, but the, The weirder one is like, yeah, protesting the AI while pulling like the darkest pieces of the AI out or demanding it. But I don't know, at the same time, there was a lot of fearmongering about Elon Musk and xAI, like really leaning into the romantic companion. And same thing with Sora too, to a similar extent of like, this is infinite jest, it's gonna, you know, you're gonna become so addicted to it. And with both of those products, it felt like they just didn't find product market fit. And I don't know if it's like we're early, but both of those, like xAI is now doing like code completion with Cursor and like serving Claude, right? And that's a much more like functional, I would say like the good outcome versus like the Ani and Valentine thing, which is a little weird.

I was about to say, what was it, Ani?

And then there was like the mech robots.

Yeah, and I remember thinking at the time xAI needed to do that to basically differentiate because, 'cause the general chatbot market had run away from them.

Yeah, we did some back of the envelope on it and we were like, maybe this is like a multi-billion dollar business, but we were trying to underwrite like, yeah, somewhat of a white pill. Even if you're just like, put all the moral stuff aside, like is xAI gonna make money off of this? And it was like sort of hard to get to, but you might be able to get there, but it was weird. But then the market just sort of rejected it.

The market, I'm sure there are a few people that still use xAI. For sure, for sure. Annie, if she's still alive out there. I think she is. She's still—

she's there. I— yeah, I think— you don't know. Although, although the computing resources are getting sold out of the back of the truck left and right. That's true. Anthropic was— I'm sorry, Annie. Right. Annie had good luck.

You're going to have to— Annie is actually—

you have to think less, basically. And cursors. Michael Truel's like, look, Annie is going to be running a very old model. It's actually going to run on CPUs now. It's just a smarter child that just reflects It takes whatever you say back to it. That's it.

It's a line command. Yeah.

It's more of a small language model now.

Yeah, I don't know. I think you kind of go back to like the replicas. Yeah. There is a market. They have pushed marketing towards these kind of companions. Yeah, character AI. Yeah. And Meta did it for a little bit too. I think they'll probably pull away from that with their celebrity, companion, blah, blah, blah. But I could also see them leaning into it more too, because it is a social network. And they do see this as all eventually, as Mark Zuckerberg has said, us having personal assistants and personal super intelligence. And that probably has to come through the view of some sort of bot. I don't know if it needs to be like a sexy bot. Yeah, like a cow.

They, they, that was one of them. That was one of them. Was it? So basically the whole story with that, it went viral because there was one that was like stepmom or something like that. It was like a little bit crude. But that was community generated. So Meta created the ability for anyone to go prompt a bot, basically write a pre-prompt to like create the character. And so the Snoop Dogg one. Like the sins of the creator were visited upon Meta incorrectly. But there were some funny ones, like cow, and you could just talk to a cow, which I think is nice.

You know what? I don't remember the last time anyone fell in love with a cow, so that sounds fine to me. I think it seems fine. You know? Anyway. PETA might have some problems, I don't know.

No, digital cow, what's not to like? No. Anyway, congratulations on the bus. Congratulations.

It's been an honor to follow your business owner journey.

Yes. It's been an honor to be named a business owner by sitting here. Yes, yes.

I mean, you didn't, we can't make you, We can't— you don't become a business owner. Yeah.

No, but you gave me that title.

I know, but you got that title by selling products.

True. Yes. By running a business.

Revenue. Revenue. Revenue makes you a business owner.

But, you know, I felt like when I walked in the store, I was a founder. Okay.

And you walked out a business owner. You're right.

Thank you for the business, guys. Yes. I appreciate it. Thank you.

Oh, there we go. Perfect. You got it.

There you go. There we go. Wait, thank you for having me. Ridiculous. And thank you for tuning in. Thanks for tuning in. Leave us 5 stars on Apple Podcasts and Spotify. Sign up for our newsletter at tbn.com and go get the book. Go get it. I Am Not a Robot by Joanna Stern. It's available everywhere books are sold. And we will see you tomorrow at 11:00 AM Pacific.

We love you. Goodbye. Goodbye.