The AI Revolution Call: The future of AI commerce!

You know.

In this day and age, every TV show should have a unique weekly intro AI-generated video. That stuff is hilarious, man. The USS Willard.

Yeah, I wanted to do a special space one for the Artemis launch or the Artemis flyby that we had today.

Yeah, that was awesome. That was great. Was it a shuttle, Bryce? Is that the only complaint we can have?

I wasn't sure that it was actually a shuttle. It sort of looked like a merger between a shuttle and a starship.

Yeah. So that's excellent, Craig.

Well done.

Welcome, everybody.

Where—

what happened to Johnny?

I don't know. He's taking care of the pup.

He was upset by the intro video.

No, that's, um, not intentional. Somehow I lost my, my video as I was making a switch to to, uh, make all my screens show up. I have a, I have a switching device that goes between— I'll get it in a second.

I wrote an article we haven't published yet, but I sent it to Bryce, and I'm hitting mute until you got her figured out there, Johnny Boy. Anyway, let me do the quick intro. Well, there's Johnny. Welcome everybody. Oh, I unmute. I did unmute. There we go. The AI Revolution Call. I'm Cody Willard. That's Bryce Smith, Craig DeLong. We work at 10,000 Day Funds. John Dio is an AI expert, 30 years of experience developing, learning, perfecting machine learning and AI. Um, the, uh, I did an article. I haven't published it yet, but it's about the fact that the whole promise of AI is that it should be easy to use for anybody. There should be no learning curve to AI. The learning curve is talking succinctly, unlike me right now. Um, it's prompting well, effective communication skills. And you should be able soon, like within months, to be like, Hey computer, put me in the StreamYard call. Make sure it's on my Facebook and YouTube and Twitter. And it'll be like, you mean X? Yes, of course, why even bother asking me that? And then you'll be like, hey computer, I'm on the call but my, my camera just switched off. Can you fix that? Like, all of this stuff is very simple. It— now that AI exists. It's very low-hanging fruit for us to change any setting, to interface with people, computers, cloud apps, whatever it is, by just simply talking. You could— you can type on this archaic, soon-to-be doomed device that interfaces with the cloud and the computer.

Yeah, it's like the same one.

Not quite. No, this is Apple.

Oh, okay.

Great. We were in New York City 3 weeks ago, and, uh, my friend Paul, who runs a giant hedge fund, used to work for, uh, Ray Dalio for many years. He had— no, no, that was wrong. I got it. My wires crossed. Zach, my friend who runs a very successful short-selling hedge fund, had a keyboard on his table. Why did you like it again?

Well, it's the Model M, and the Model M keyboard from IBM was like the best thing IBM has ever done.

Tell John what you're referencing here?

1985.

So he literally has keyboards that someone he bought on Facebook garage sales, and it's someone dumpster dived at a corporate office for the keyboards when they were throwing them out. And Zach has like 8 of them because it's apparently the greatest feeling, sensors, keyboard sounds, all of it. Pre took pictures, he was so excited. I I'm in there looking at Zach's like spreadsheet of SaaS companies, every SaaS company worth more than a billion dollars and how he could fit them into different parameters. And Craig was over there taking pictures of the keyboard because he was so excited.

Yeah, I couldn't pay attention to like what I was—

Zach was wildly proud. He was like, oh my gosh, somebody appreciates my keyboard. Yes, look.

Um, Craig's still occupied with 1985.

Yes, I am.

Yeah.

Song there. You didn't know the reference?

That was a Suits reference.

Bowling for Soup. It's a song. It's called 1985. She's still preoccupied with 1985.

No, it's pretty good.

Ask AI to help me with that joke later, Bryce. Anyway, the point is that It's, it's pretty remarkable that this day and age there's still a learning curve to get on a Zoom call, to get on a live stream, to say like log on to your computer. Like it is going to be all spoken and/or thought within 5 or 10 years. A lot of people are going to choose to put a chip in their ear or get it permanently implanted, temporarily implanted, I guess.

Could be. Well, I've seen reports of just the new laptops, people wanting Edge AI on new laptops, and the obvious transition to that is just talking to your laptop. Hey, do this for me. Open up this—

It's at the point now, John, correct me if I'm wrong, the technology clearly already exists. It just hasn't been implemented that Siri can't do shit. Like, you should be able to already just be like, hey Siri, uh, change the setting so it's not so bright.

Siri is just so disappointing.

All right, but Gemini, fine, Gemini. And Siri will soon thereafter follow because they're going to build it using Gemini, which is owned by Google. But yeah, like, literally we are at a stage where you talk to your car. If you have a Tesla that was built within the last 5 years, you're rolling robot, and it now has Grok AI built into it, and you can debate the best route before it puts it in the car. And that's coming to your smartphone. It's insane that Tesla's car is better than Apple's iPhone.

That's, that's a great statement, right?

The AI in Tesla's car is infinitely better than Apple Siri. Bryce, you agree? Like, you can talk to your Tesla and it actually answers questions and implements instructions.

For sure.

Some shit. And it's like, I don't know why I cuss on these.

Siri's not even at— what's that?

You have to wonder, like, why even Google has Gemini. Like, I had an Android phone up till about a month ago. Like, it The Gemini on Android could not manipulate settings on the phone. Like, and it could do a lot of things. I see in beta now, you can integrate with apps directly. So like, you could be like, hey, Gemini, order my favorite cup from Starbucks, and it will do that. And I just, I just, you have to wonder why it hasn't happened faster.

Like, Bryce, I'll almost guarantee— sorry, Johnny, we're coming back to you. But I'll almost guarantee that my Tesla app on my smartphone will have Grok capabilities on it. And I can say, hey, Tesla, turn on dog mode in my car. And it'll do that, change that setting and turn on dog mode to keep my dogs cool before Siri is going to be able to turn the light, the brightness down.

What I'm saying, Cody, I guess, is why you have— you should ask yourself why that hasn't happened yet, is what I'm saying. Because I feel like the technology exists.

Siri is—

why hasn't it?

That's what I'm saying. We should ask that question. Johnny, what do you think, John?

Siri's not an LLM, is it?

No, it's trying to figure out how to tie into an LLM and hasn't done it well.

It's a generation behind, or several, where you can't make—

but he's talking about even Gemini.

He—

that's what he's saying. He had a Google Gemini phone, the latest version, up until a month ago. And he couldn't say, hey Gemini, turn down the brightness.

Well, it was also that was every LLM was until very recently a chat in a box, not a language action model, which is where you're headed with these things, computer use. Anthropic has computer use, not perfect, but it can go and do stuff with your you know, computer, um, get it to do things kind of like Manus.

So Craig, Bryce, I think the answer to your question is we're being impatient. I mean, the AI revolution is happening faster than any revolution in history, and it just takes a few more months before we can tell our computer and our phone to change settings.

I guess what I'm thinking is that like, um, this new— the reason that all of the apps now can accomplish things for the most part is because of Model Context Protocol. So the apps can talk to the— the AIs can talk to other apps natively on the backend and accomplish things. And it makes it smooth and reliable. You know, you can do it with APIs, but it's not quite as good as MCP servers. So do you need something on the backend, like within, within the iPhone, John? Do they have to build some kind of a uniform protocol or MCP, or just adopt MCP so that Siri or Gemini can talk to all apps that are built on the platform natively, including settings.

There are different layers. So there's, you know, APIs first off, direct kind of using something that was hooked in to work that way. MCP, it comes, you know, next. And then if you don't have MCP, the computer use has the thing where it'll take a picture of the screen and know where the mouse should click and do the click for you. So it should be able to do it even with, uh, even with machines that are not. And now there's, uh, is it called MCP CLI or what? It's, it's a, um, it's this the CLI aspect where you instead of an MCP and you have to program that into position, it's like when you make new software, make it that it's, um, that you can do things from the command line And then whatever you could do from the command line can be done by the, like, remote controlled by, by the LLM.

John, how far, how long is it before the engineers at Apple can speak to an AI thingy, agent, LLM, whatever it is, and tell it, hey, we need the capability for that what you just described. We need NCP and CLI and we need it done so that people can change the brightness setting by telling Gemini. Like, why does someone have to—

what?

Can't the LLM, can't the AI coding agent do it?

Yeah. So there's stuff it does already. Like, I use a Mac and I am sitting here working with Claude doing programming and I said, I'm tired of having to keep open a terminal window by choosing that application and running it and then getting it to go to a certain place and run something and it says, oh yeah, okay, I'll do that for you. It does it now. It made it into a shell command, which is fine, that's what I wanted. But it made it so that when I just say this one thing, it launches that application, switches to the right place and runs the right thing. Which is essentially like doing that, doing it with a particular application. But like other applications can do that too. Apple has their built-in, um, what, what do you ever call it? It used to be AppleScript. I forget what they call it now. Um, but their, their way of doing kind of macros on running things in applications, and it just needs to be, you know, you can interface with that. And the, the biggest hesitation I saw over a couple months ago that SARS starting to go away, I think, is this fear. Oh no, what's going to happen if it grabs the wrong stuff off your machine or deletes things or whatever? So they're, you know, getting past the whole, you know, is my tanker insured to go through the Strait? No, like, can I actually, um, you know, uh, go and do this on my computer safely? And because that it's capable already.

That's just it. I mean, In some ways, Apple is like this legacy, backward-looking, slow-moving entity that is just taking time to get it put in, in a good way. Like, and what's crazy is Apple is so dominant on their iPhone platform specifically. That they're 3 or 4 years behind Gemini and it doesn't even matter. I'm still locked in, still at my smartphone, my iPhone right over there.

Speaking of Apple, I wanted to, uh, bring up CarPlay because for non-Tesla drivers, CarPlay is still just kind of, kind of garbage. It's not—

it— okay, good. I didn't know where you were going with that. I was about to interject and be like, the worst thing ever.

Yeah, it's pretty bad.

It's awful.

It cuts out all the time.

It just even like, it's so annoying when you're like, all right, let me put this map on my phone. Oh, let me plug it in. And then immediately, because I got to charge it, and then immediately you've lost the map on the phone because it wants to put it on the crappy screen inside your Toyota.

Yeah. And then OpenAI just announced it's going to be on CarPlay. And I asked that, I looked into it. I was like, so is OpenAI going to be able, I mean, Is there any type of thing that it can actually do other than talk to it? And it's like, no, you actually have to hit the icon, press it, and then just have a conversation with your AI. It's like you could already do that. Like, if I just had my phone in my hand, I just press the voice button.

But the reason why it is something is because it is more distribution for OpenAI. I was excited when, for example, Spotify was included on the Peloton.

Yeah.

Because it's just more distribution, critical mass. So it's not a bad thing for OpenAI, but you're exactly right. Like, it's lame. Yeah.

Yes, that's my point. It's not bad because distribution is good, but it is not cool.

So Cody, this is like more of a business question, I guess, is Apple has been around for X bajillion years.

50 years as of last Friday, John. How do you not know that?

Uh, I was— I, I had to calculate here. I could have said about 40, 49, but okay. So, uh, the Apple IIe in 1977. So, um, the, the, um, they've been around all this time, especially the Mac has been around since around 1990 or 80-something. But, um, they've got so many people there, they're so invested in doing what they already do that the longer they get like that, the harder it is to change anything. And there is— they should obviously, with the hundreds of billions of dollars that they have and very smart team of people, they should be able to say, hey, let's go do this thing. But they are risk-averse, uh, on a very large scale, and it's built in.

Tim Cook, the CEO, is like the classic example of risk-averse. He just wants to make sure he doesn't destroy what Steve Jobs built. Bryce, you know, on that same note of thinking this stuff through, one of the things I wrote in that piece I sent you guys last night was that apps, like the AI platforms, subsume the role of the apps instead of empowering apps. And last week on this show, I showed you guys something I just made with Manus over the weekend, and it was an AI app discovery. TikTok-like website app thingy. And after I finished my article last night, I was like, well, that was a silly endeavor to mess around wasting time with that, because the whole point of like, like the AI apps, most of like, you could speak to— if not today, within months or a year, you'll simply speak to Gemini and be like, make me this game, make me this app that I have in my mind that will create like skill challenges for doing scales on a guitar. My daughter was playing with an app that does that the other day. And like Manus could do that right now. I can prompt it, take 45 minutes or a weekend to make the app, but I could totally do that. And that's where I'm finding this interesting, what you just said, John, in the context of Apple. To this moment, even Elon would tell you that Apple's value is the discounted cash flow of the App Store. Is that right, Bryce? That's the quote you always use. You're on mute because we— you were typing.

Yeah.

I mean the net present value of it, but that's what I meant. Yeah.

NPV of the app. Of the Apple App Store. And I'm not sure that's true anymore. But if it were true, Apple stock would be doing what Microsoft and the SaaS stocks are doing. It would be cut in half or more by now because, like, we don't— like, I literally was— one of the things that made me think about writing the thing I wrote this weekend was I was sitting there with my daughter doing this scales competition on our guitars, and it was hilarious. And she was so into it and had to beat her time and then beat her time. And I mean, normally I have to be like, hey honey, make sure you practice guitar for 35 minutes tonight. She was like Saturday morning bringing her iPad and her guitar in like, Dad, can we do this? Nah, we're competing. And it's hilarious. And I got to thinking later, I'm like, that's an incredible app, but there were none of the songs that I know. I would have liked to have other features. And then I thought, I bet I could build that guitar teaching app by the end of this weekend. I'm not— I didn't, but Manus could do it. It burned you a lot of credit, several hundred dollars worth probably, but you could build apps. And it's getting to a point where, correct me if I'm wrong, John, Bryce, Craig, within a year or three, you're simply going to open your iPhone go to Gemini and say, make me a skills competition for the scale of G on a guitar and listen to the notes and make sure we're doing it and time us.

No, I think it will be. I think that'll be just a part of life. And Apple has recently been putting their foot down against vibe-coded apps. But I also think that will eventually come to an end, or there'll be some type of cottage industry that's like, all right, we'll stamp your app, or maybe Apple will do it themselves. Like, they'll approve apps to go on their store, you pay a fee, and then you pay more fees as people use your app.

You know what you did there? That was great, Craig. The bottleneck, the constraint, is Apple's App Store approval. And if they are putting a new constraint in there, there will be a solution. It might be a third-party solution, it might be an internal solution, but they're going to do it. Bryce, your thoughts?

Yeah, I think it just depends on the app. Because what you're describing is sort of like the simple consumer app thing, which, yeah, the AI could build. But like, those aren't the biggest— the publicly traded companies are not the Guitar Lesson app companies of the world, right? And so like, it's like, okay, that happens, doesn't really move the needle for anything. And so then you have to think about like the real apps that people care about and use all the time— the Ubers, Airbnb, DoorDashes, Facebooks, things like that. Like, why do those exist and why are they so valuable? It's primarily not because of their user interface or what they do on your device, right? DoorDash— I could vibe code the DoorDash UI right now, right?

But you can't get restaurants to join up.

To sign up, right? I can't get drivers to sign up. And so yes, that's gonna happen, but it's like, what, what needle moving happens because of that is less, less certain, right?

John, if I may real quickly, we, we spent months, the last several months, Craig Bryce and I, trying to come up with some of the things that make applications, third-party applications specifically, valuable. And what we've always ended up centering it around is Sprice is hinting on there. It's like not just unique data, but more importantly, like unique users. Meta would be another example. You can't get a billion people off of Meta into another platform overnight, TikTok notwithstanding. It's very difficult to crack into something once it hits critical mass on both sides of creation or delivery, whatever the services and the consumers and the people that are using it. And that's a very valuable thing when you get to that critical mass. Bryce, the example you gave me this morning that is a consumer publicly traded app, Duolingo, like that is probably better than the guitar example even. You presently can right this moment open Gemini and say, make me a fun game for 10-year-olds centered around basketball to learn Spanish.

Sure.

John.

Right.

Um, so coming from, from the outside, outside of you guys, you know, going and talking about this, I have a different, uh, way of thinking of it, which is that, um, I don't, uh, I don't accept the question. You're putting the question around, uh, apps as if that's the basis for knowing what's going to go on in the future or what's significance of the advance of things. And a couple reasons I say, number one, you got— you're not the audience, you're not the, you're not the, the, you're not the, the market, you're, you're the exception. People who go and make apps when they go think of something are still a very, very small percentage of people. The other is, are apps really the kind of unit of exchange in the future, uh, you know, we see all of what's happening to, to SaaS, we say, well, stuff will still happen, it'll always still be, you know, companies doing things and everything. It's like, yes, there will. What will they be doing? What kind of the— what's the relationship that they'll have to kind of the billion users of AI?

Um, yeah, John, the first point I completely refute because yes, at this moment It is a unique thing for Bryce, Cody Willard, and Craig Smith to have. I assign them, like, literally, we immerse ourselves in whatever technology we're studying. And so it's yes, for this moment, not everybody's doing it. But that's my whole point, is that you're going to get to a point where you're not vibe coding a language app. You give it one simple prompt. And it creates it for you uniquely on the spot. You interface with it, you close it, you can tell it to— I mean, it'll save it automatically. You can go back at any time and open it back up in that thread.

But it transcends the very concept of apps, is what it does, which makes the question irrelevant because it's like, okay, so when it goes and makes the app automatically, you know, you vibe coded it without even thinking about it. You were just talking. Then you get like what's happening already with, uh, Claude, where— and several of the others did this as well— where the answer to your question is the correct—

that's exactly what I'm getting at. Maybe we're saying the same thing.

The thing about that is you're not—

even right now, last 2 weeks ago, I told Manus, another weekend project, my daughter wants a she shed. And we have a, you know, a shed that we've had sitting out by our house for 60 years. And we're like, hey, here's the size of the she shed. Like, we want to have some frame of reference for what a budget would look like and some designs. And we took a picture of it, sent it to Manus, and took 45 minutes and several hours of like interfacing and improving it. But there— we literally have an app about she sheds that shows you pricing schedules for cheap, medium, or expensive, shows you each component's cost. It's thought through whether you need to do the windows or whether you want to paint the floor, whether you're gonna— like, it's incredible. And it made it without prompting an app. Like, that's my whole point, is that— yeah, so like, and someday it will do that without the 45 minutes. It will be—

so at that point are we going to even think of it as having made an app, which it did, or is that app just a side effect of the fact that it made a she shed? That's like, right, I'm like, I asked for a she shed and I got a she shed, right? Um, you know, and I, I got to describe it to it and we talked about it and so forth, and then there was this she shed. Not, oh look, it made sliders and a window and buttons, right?

That's right. Bryce, you, you agree and you've seen what— are we saying the same thing?

I, I think so. Because I guess probably, I think what John is pushing back on is the idea that like, you won't, because applications do things, and you still want things to be done for you in the future. So it's just a matter of how the things get done. And whether that's by clicking on buttons and things or talking to the AI and having it do the thing.

There's not an economy to talk about, like at that point, there's not, you don't say like the app economy, It's just, that's just, you know, it's like, well, what is the economy then? I don't know, but maybe it's not like, well, who are the winners who did this with apps? It's like, how did apps happen?

Go ahead. All I was going to say is, in terms of the business implications of it, I think like for applications that derive majority of their revenue from advertising, for example, that's significantly a risk because you lose— the eyeballs are gone. In the— in Cody's future of how this all is all happening, the eyeballs are on the Gemini, ChatGPT, Claude app, right? Not any Expedia, Duolingo, anything like that. So it's like, if your business model is dependent on that, you're sort of screwed. If you're DoorDash and your business model is based on transaction fees from providing a service, it's like, well, it's sort of hard to think about how AI changes that.

And Bryce, it's interesting too. There's something you're getting on there. If you're a door— it's the conduit apps. It's the apps that, that bring one side of service, like delivery to the consumer. And so you need that critical mass of whatever it is that you're the distribute— you're distributing, you're distributing food to the masses. Well, I need a map that has an app that has critical mass that does that. And what's interesting is even then you're not going to be pushing the button. You're literally going to go, hey Gemini, order me McDonald's, make sure you pay for the extra time to get it. Or it won't even have to tell. It's like, I know this guy always pays for the quick delivery and it's there and it uses, it'll even check. It'll be like, should we use Uber Eats or DoorDash? Let's see which one's saying the fastest and the cheapest. Bang, done, got you. It's DoorDash, charge to your account. Bing bong, hey, I got a Big Mac.

So the thing that I'm saying that I think is, is different is, well, one of the things is that there are people who think in terms of apps and, and, and where the outcome is, uh, like, hey, go make me this thing that'll do this chord competition, or go make me this thing that'll make this shed, or whatever. And that's not most people. And in the future, it'll be maybe even fewer because it's never, it's never been about that. That's just this thing that had to be there in order to make something happen.

Yeah. You're literally, you won't prompt it as I did a minute ago about the guitar teaching, that, that, that, that, you'll just say, hey, my daughter wants to learn guitar. She's 10 years old.

Right.

And it'll do the rest. But I, John, actually, I think we are actually saying the same thing because what my premise and what started us down this rabbit hole of apps versus non-apps is that AI, like, is the platform and is the app. It's like the— what the internet did, what computers did, what smartphones did. They— they were platforms. And yes, Microsoft had Windows operating system, and it also sold you Live 3 or Windows, uh, Word 360 and all that, Excel and all that BS that doesn't— you don't pay for anymore. And it was— but it was like you could download a video game to play on that platform. And in the future, sure, you'll still have video games. I'm not saying it's all or nothing, it's— but increasingly you're going to sit down somewhere in the next few years and instead of turning on a PlayStation 5 and passively opening up that game and playing somebody's vision of what is a cool game. And yes, that'll still be cool. You want that. Artists will make great games still, but you'll be able to be like, hey, I want to be a Jedi Knight and I would actually like to be in prehistoric Earth.

So here's a thought experiment. If, um, If OpenAI, or whoever it is, right, the, the platform is the iPhone, then what's the App Store?

That's my point too. There is no value to the App Store.

No, no, it's— you gotta like back off of that metaphor a little bit. It's not literally an App Store, but like, isn't it possible that some like either famous musician, or, you know, somebody else who is musical but also a great teacher, whatever, that there's a market for the intelligence of these people to plug in and to provide, you know, good— they tried it, um, ChatGPT tried it right at the beginning with, with their plugins, but that was very basic, very early on. But now with agents and everything, if you could go make an agent and it's like, I know I've got the job—

100% agree with you, it is not mutually exclusive. I, that It's bigger than one thing. A lot of people are just going to go, hey, chat Gemini, my daughter's 10-year-old wants to learn guitar, go. And other people are going to be like, hey, you know, Bob Weir is my favorite guitarist. I would like to learn guitar in the style of Bob Weir. And Bob Weir could have, you know, God rest his soul, if he were still around, or his family now could be like, hey, let's license Bob Weir's musical style to ChatGPT and Gemini. So when someone asks, they can teach them in that style legally.

You know, Cody Willard, what I'm kind of getting from this conversation is it's almost like the AI is going to be prompting you. It's like, I will build this for you, like, right now. Like, you like this? Like, it tells you right now, do you want to know more about this? Well, in the future, when tokens are, you know, cost basically nothing, it's like, I can just build this for you right now. And we've seen that with Manus, you know, I think Manus is like, the best way to show this information is through a website, you know, and it, it'll just build it for you.

And there's a few different things like that that I think we're all on the same page. Exactly. I think this is what we're getting at is that it sort of subsumes it. It transcends the concept of what we think of apps are right now. AI is the app. John, the biggest takeaway from what I wrote is that Compute is the only constraint now. Like, there is nothing that AI— there is no app that you can imagine that AI couldn't build for you, at least the software side of it.

Yeah, yeah, that's what I was— yeah, so this is the, um, bits versus atoms thing. You know, at the beginning of this call, my display wasn't working, and it wasn't because, oh, how do you make this happen? It's because I've got a piece of a, a switch that doesn't work great. That I use to quit, uh, move between machines and, and, and monitor and so forth, and it gets a little heat going on it and it's go— and it's trouble. And the point is, there's a whole physical world beyond the computer, and the first thing we think about is the, you know, Optimus the robot. It's like, it's in the physical world. Yeah, it's in the physical world, but it's still having to deal with the everything that's not it, you know, the refrigerator and the, and the the table and the coffee machine and everything that's not it is a dumb thing in the physical world that both Optimus and the AI on the computer together ought to be able to start making, like remaking the whole world in, in, in kind of in their image, you know.

Bryce, this sits on our top. Were you about to say something anyway?

Well, I guess my, my only pushback is I've just been reasoning through this is just that It's like networks and marketplaces and things like that. No amount of compute can brute force its way to create that. So it's like, if you just say compute is like compute's the ultimate, I think what he said is like compute's the only constraint to building any app you want. It can't do eBay, it can't do Facebook Marketplace. So it's like, there is, there is something else there.

Besides, in certain cases you can do Duolingo, but you can't do something that has—

so that's how do you— how would you change your sentence? And it's because the sentence that the, the computer is the only restriction on building any app is not true. It can't build eBay.

It can build eBay, the app.

You can build eBay, but it won't succeed. I guess that's the point.

Well, there's a difference between building a hammer and building the next Craftsman, right? Like, you, you can build your tool that you need, but as far as like having the vision, laying the groundwork, and creating a business. That's—

eBay's a company.

The human element.

eBay the app, it can build. eBay the company is a two-sided marketplace, and it can't fill up the two sides.

Again, I feel— but you know what's crazy, Bryce, is I don't think it's— I mean, I don't think it was a week after the first ChatGPT moment that we started trying to triangulate what are the What are the elements of apps that are not replicable in coding? I mean, maybe it wasn't the first week, but you know what I'm saying? We've been trying to think through the Uber discussion versus the Duolingo discussion for many, many months. And it feels like I've gone down this whole rabbit hole and ended up right back where we sort of were with our fundamental premise that if you don't have critical mass of data or suppliers or users, you need all of that, then it like, yes, you can make the app, but it doesn't even matter. eBay or Meta is probably a better example, will still beat you all day every day.

If you build it, they will come is not true, but it's possible because ChatGPT, They built a thing and they had the most users the quickest of anything ever in the world before. You know, like you can have a large set of users, but you got to make something great.

Yeah. So Bryce and I had this discussion earlier today. We didn't call it atoms versus electrons, but it's— I guess what we ended up with was the electron as like the bits, the issue. Well, the electrons, literally, we were like, If you're these, even if you're designing chips, you're Nvidia, you're these, you're, you're not actually in the atoms world. You are simply building bits and bytes that design how electrons are moved. And then you send that to Taiwan Semiconductor and they make, they use atoms to make the chips. And so it's just sort of been an interesting exercise even today, realizing that On a very fundamental level, NVIDIA is a software company and not a hardware company.

And then even the hardware companies are probably 90% software.

Like who?

TSMC. Like, yeah, because everything about their process and design for the machines that make the stuff You know, it's, it's not, you know, before you get down to the, oh look, it's actually doing this with lithography, it's like, yeah, okay, that, that happens. But all of the work, you could, you can figure that based on like the number of employees and where they work, how many of them are in bunny suits and how many of them are sitting at, uh, computers behind that.

Um, Bryce, this is a really fascinating discussion because in my mind TSMC would be 90% hardware, 10% software. But there's an argument— I mean, it gets to be semantics in some ways, but like, there's an argument to be made that, that John's point is, is valid, because even at TSMC, most of the guys are not physically etching lithography, right?

I mean, but that's, that's true of almost all modern manufacturing for the most part. ASML would be the same, right? It's a hard— ASML is the definition of a hardcore company, and it's a bunch of guys at computers sitting around making designs on electrons and bytes and bits, right?

Every company is using software to, let's say, to create, you know, it's a tool, you know. So I mean, I agree with that.

Can AI— like, like, I'm just thinking, I'm, I'm stuck on the idea of like the networks and things. Like, if, if eBay, like, didn't exist, let's say, and you're just like, I want a 19— like an original Gibson, whatever, guitar, and eBay doesn't exist, and some other guy's like, yeah, I've got one I want to sell, like, the LLM still— AGI still can't solve that, right? Like, what does it do? How does it get— how does AI get you the Gibson guitar you want and sell and get the money to the other guy if there's not a network for that to happen on.

You almost just— you almost just talked me into the— like, I think you could code— you could make an eBay, and then you don't even have to get the suppliers. You just got to get the users. If you can get users on your new eBay and all it does is search the web for guitar, like whatever product you're looking for, it's sort of You have to have a personal website where you can list things, and then the AI— like, it could go on to Alibaba, eBay, Amazon, Walmart, anywhere that's got a secondary market selling place.

Facebook Marketplace.

Facebook Marketplace for you. Yeah.

What's the actual, um, kind of limiting factor? You know, you've got— for one thing is Oh, well, they have a payment system. Oh, that's easy to get separate from eBay. So you can get a thing, you could get, uh, what's the, the stuff that people use all the time now for just paying through, uh, through the email with each other?

Um, I don't know, Zelle or something.

Or Zelle, like, like, yeah, like a Zelle kind of a thing. I mean, it's like making payments happen, that's not a problem. Oh, but what about the advertising? That's all there. Okay, so there aren't other platforms that get your stuff seen by lots of people.

I guess what I'm almost picturing is that like, we're so— the internet as it existed with websites and things is so fundamentally baked in. You can't really picture a world without it. But I'm almost saying like, what if your AI could just like— you could like make a public facing of you. And you're like, here's me, and I like these foods. I like— I have this stuff for sale. I would like to buy these things. And then those in a world that's not www., then like the other Geminis of the world that belong to people like, yes, this guy has this and he's told us he has it.

You're getting it, Bryce.

You see what I'm saying?

Yes, 100%. To this day, like what data, what Magnus goes and searches is content that was created to tell human beings something. And there certainly will just simply be a future where your AI in your lapel or in your ear or your glasses is with you all the time. And you're like, you walk into the room, the storage room back there, and you're like, you know, I got those 3 guitars. I would like to sell that, that one in the middle.

And it's like, even if you didn't have that thought, if your AI sees everything you see And then it, and then it sees out somewhere else, like they're looking for your, uh, Martin whatever rosewood guitar. Hey, I know you have one. This guy's interested in buying one for $1,200. Would you be willing to sell it? And you're like, yeah, I'd do that.

So you're doing some disintermediation with all the stuff like eBay and everything. And if you go, well, I'm going to tell my AI, and I know that on the other end all those people running their, you know, open clause or, or their manises, you know, are some of the— some of them are interested in, uh, getting a guitar. And in between, there's— even before it exists as a thing, there's like 20 or 30 possible ways and things to be using. And both sides know about all those things in between, and they're like, I can go crawl through all that.

And that's where Google and Apple have bigger edge than everyone else because they do know what's in your emails. They do know that you purchased something a year ago and now someone else wants it and the price has gone up and they know that it might be interesting to you. And it can do all of this even already. So much of what we're discussing goes back to where I'm like, it's just a compute constraint. Like the future, everything we're discussing could be here today, but we don't have the amount of compute to make it happen.

I feel like that's what I was getting at. I like—

I—

okay, there's something true about your thing about compute solves everything, and it's like, if, if everyone has this personal AI, like, then that's possible. You don't even need eBay. And to get personal AI like that, you just need more compute.

Take it to the even bigger extreme, like Uber. Literally, there could— this seems totally outlandish and like like 20 years into a Harrison Ford dystopian movie or something. But couldn't it like— it's like, hey, you know, your robo taxi sitting idle and that guy over there needs a ride. Do you want to let it go?

Right.

You want to go get them?

Right.

And it's like it disintermediates even Uber. Like, you don't need drivers. You don't need customers. You just, hey, I'm on Gemini and I'm on ChatGPT and it knows where the cars are and Right.

Mark Zuckerberg has said that the business will have an AI just like it has a website, and it will have— and it has an email. And I almost like, I almost think there's like a future where the web as we know it is not really a thing, right? There aren't.coms, there aren't websites. There will be a network of connected people. But it might like— there's so much disintermediation that is potentially available.

Craig, you go.

That's all I was gonna say. Like, the future is just that the future— maybe disintermediation is the future, is what we're saying.

John Craig, did you have something?

Well, I just thought it was, you know, funny when we think about just commerce and almost advertising. You could choose to advertise yourself, you know, you're going into like a huge, like, Superdome— like, with Superdome's in New Orleans, but like, you know, a concert, and you have a car for sale, and you could just basically tell your AI, hey, just put this in my whatever profile, and people could see that I have this car for sale when I go into this giant thing. It's like how you're riding down the highway with a for sale sign, but you can choose to have that up or not. And it just, it almost changes the way advertising is done on the, on the smaller level. And I, it's It's just kind of fascinating to me when you think about the new internet and the way that everything's ones and zeros, and it'll just happen like that.

And it, you know, like you see on the side of the road a car and it'll have a for sale sign on it. It's just sitting out there, whatever. Now instead of that, they'll just have on Zuckerberg's glasses and you'll look over and it have an overlay, right?

Yeah.

But it is because those, the information comes together. You know, but the, um, so I, I, uh, a friend who just the other day had, uh, a bad experience with both, uh, Tesla and Amazon. Not the Tesla with the car, but with like getting a Tesla product. And, uh, it was like just like the Siri thing you're talking about. It's like, well, if AI is where it is, in both cases it was, uh, they were in touch with, uh, the, the place and actually through chat and it was the AI that was doing the chat, and it was just incompetent and saying stupid stuff and like throwing them off. It's like, why? If that is the best use case for— and it's really one of the best use cases for AI right now to make that stuff way better— why is it such garbage?

Great question. It's because it takes time to get implemented and improved and to get the good version out to everyone.

It takes time because the missing data I think there's a fundamental—

I think the reason the network thing and marketplace thing is such a big deal, and what I realize is like, this disintermediation thing is a huge threat to lots and lots of businesses, mostly, most of which are applications on the phones. And if they, if they open up their ecosystem and say iOS or Gemini Android, can like now use our pipes to get these things done, then they've lost the user, right? Like for Expedia, let's say their whole model is built on like coming to the app, they get eyeballs, they get advertising. If they all of a sudden say, Gemini, you can come into this app and place for a hotel based on what the user said in the Gemini app, like that's sort of an existential threat, threat to like 90% of Expedia's revenue.

Yeah. And Bryce, I think it's—

they won't do that, right?

Amazon right now is told— I don't remember which one— what, maybe it was all of them— it's like, no, you can't scrape, right? Or you can't scrape our store. You can't, you can't. And it's exactly— I think that they've got an inkling that that's an existential threat to let them do that. What's interesting is Amazon, for the most part, is distributing products made by other places, of course, but it is the conduit for those products. And it's the one layer to get to them. It's one markup, you know, I'm sure the distribution to the guys who send it over in the container from China are getting a markup, et cetera, but it's one markup for, to, from the actual physical store. Expedia makes a markup on the data that it gets from American Airlines. The data for Expedia is not its own. It is simply collecting third-party data from its partner airlines and hotels. And that information is on—

so, uh, that sort of answers my question, and because I'm like, Expedia will never open the closed garden, right? That would be totally irrational for them. So how does this get resolved? Then the hotels go straight— they're like, you come, Gemini, you can come to Marriott.

Gemini would make— you're Marriott, you would love that. Because look, Gemini maybe someday is going to figure out a way to mark that up somehow and get a cut of the revenue. To, but it might be 2 or 3%. It's not going to be a 20 or 30 or 50% markup like what Marriott has to suffer through when Expedia buys a bunch of rooms for discounts.

That's fucking interesting, man. It is interesting, right? Like, it's like, to me, your, your first layer of thought is that these marketplaces are exceptionally well positioned to survive because they're not just a UI, right? But then you're like, they're, they're just middlemen that are marking up somebody else's product, right? And if, like, especially you can see it at the enterprise level, it's not that hard for Marriott and Hilton to go straight to Google and ChatGPT and like, here's our inventory, search it, you get 5% or whatever it is, and not pay Expedia.

I mean, holy crap, this is really enlightening. This has been a really good discussion today because I think I made sort of a couple of epiphanies about— yeah, like Amazon is not as threat— Amazon store itself is not threatened in the way that Expedia is threatened. Uber, not threatened in the way that Duolingo is threatened. But frankly, there is threat to all of them because Amazon store itself, not the cloud services and AI stuff, but its store itself is a middleman marking shit up. And if the Chinese vendors can eliminate that, if it's the constraint and it's a cost, an added cost, all of us consumers and all of the guys who actually make the junk that you bought on Amazon from China— old-fashioned calculator, that sort of It's not the best example. Is like, why wouldn't the AI just go around Amazon and order it straight from China's distributor and find a warehouse, an alternative to—

and that's what I mean. To me, it seems like there has to be like almost a new interface whereby those independent sellers can get connected with independent buyers. Like, the AIs can technically do it, but how do the AIs know what Shenzhen Manufacturing Corporation has available for sale?

As it is right now, they would— Manus would go out and search Shenzhen's website and hope that the data is up to speed and stuff. but it would—

that's what, like, they're almost like a new layer of the web.

At this moment, it's still searching content and websites that were created for human eyes, right? And they were like— that's a major layer of cost in the compute world, frankly. And if it's just the data of what's in the stuff and where it is, if it's just the spreadsheets and not the beauty interface on top that's made for men to—

and women to look at, It almost makes me wonder if there's going to be some type of rating review system that'll be for the AIs itself when they go out. And because if you're just buying from like some random, you know, independent seller because they happen to have what you want, in a world of like Temus and people buying stuff from random sellers and they get crap in a box that doesn't work, you want to make sure what you're buying is actually good. So how do you—

gonna happen inside the AI too though, Craig? It'll just be like 'Oh, this guy hates the Temu shit.

I got to make sure I buy him Tiffany.' But even like reviews, human reviews are a proxy for quality. They're not an objective measure of quality. And like the AI, if it was smart enough, could just reason about the product itself and tell you if it's going to work or not.

That's not always true because you could— there's literal frauds that will say this is this and then sell you something. Whole channels on Instagram that are whole— that their whole shtick is putting up that kind of content.

Although the AI is going to triangulate that from user reviews themselves, I guess, at that world.

Okay, go ahead, John.

So just— we're saying the AI is going to do it all. And I'm thinking, you know, it's not going to go bake it into its pre-trained model. When we're talking about data like, you know, how reliable is this vendor or how many people gave a 5-star for this or whatever. So there's still data and that data happens in a couple of levels. There is the pre-trained model and that's getting smarter over time, but it's also gaining in capability. But it's the capability that enables it to do, to make the Manus of the world, you know, which is that layer around it, the the harness that in, and in that harness a lot of the data will live. That's where a lot of the real kind of, where we're getting a lot of the gain in intelligence toward AGI is not down in the pre-trained piece, it's in that how it gets better at that surrounding piece. And then you have the data beyond that. Now, always been who owns the data, you know, matters. The reason Amazon is who they are, the reason Uber is who they are, because the data that they control, and they're going to lose some of that control with regard to them being the one who puts that data in front of those eyeballs. But there's still like, that data is not in the AI. It's, there's an AI that's going and making use of, of layers of data. And, you know, the question of where that all is gets back to, you know, what Cody Willard was getting into for, you know, disintermediation and, um, you know, how these different parts come together.

Last words. Johnny, you're first. Saving the best for last.

The thing about hardware versus software, I just have this, um, friend who's a hardware engineer, and so he makes, um, CPUs. Like, it was the— there was like a 386 clone, like, right down into the registers and all, whatever. And so I pictured that the way that gets made It's like, it's still like somebody sitting there doing it, but they've got like design, like CAD/CAM type laying out what's where. No, it's— there's a, there's a programming language which is the language for where does this register go. And so you're, even when you're making the lowest of low-level chip design, you're still, you're coding it. Software all the way down.

Synopsis, anyone? Craig DeLong.

This morning, I read an article about the Meta glasses, and the angle that the author took was that it's going to lead to eating disorders and all this and just unhealthy people. And I refute that on two levels. When restaurants started putting their calorie counts on the menus, it didn't just lead to rampant people just not wanting to eat anything. People still ate what they wanted to eat. Secondly, in a world where America is almost, what, 40% obese at one point, and it's steadily going down at smaller, lower, higher increments, but we need all the tools we can get to be a healthy country. And if people can have these AI tools that says, Hey, this is what you ate today, because people don't realize, like, with the amount of food that we have in front of us, you can eat 3,500 calories and not even break.

I don't think you set that up very well. So the article about Meta's glasses were that it would track what people were eating.

Yes.

Sorry. I didn't, I didn't say that.

And therefore this guy was saying that would make people have eating disorders.

Gay.

Oh, get eating. Okay. Okay. Not this. Okay. Like, it's going to make you paranoid. You're like, oh, crap.

I don't want to eat anything. Yeah.

No, it's all like, that would be— that would mean that like anyone that has a conscience and regrets in their lives is going to— there are times even just yesterday, I had too much roast beef. I went back for a second serving. And afterward, I was like, I knew I had already eaten all that. Why did I eat the second dish? That's okay.

I'm not—

I'm not going to starve myself or become bulimic today because I ate too much. I'm agreeing with you. That is nonsense. I think more information is always better. If you're going to tell someone what they're doing and with data, it's better than just letting them guess and be like— truly, the whole point is that it's going to help people eat a little bit less and more healthy. And not be insane because they're imagining shit that's not true. They're like, I know I had 3,200 calories today and I was only supposed to have 2,500 or whatever it might be. You're going to be like, ah, well, I'll forgo that ice cream tonight.

And when you go to the Chinese buffet, all you have to do is this.

Ding.

That's it. That's all you got to do. I think it's ironic that they would be talking about eating disorders coming from that when it's well documented that social media causes or increases eating disorders, both Instagram and TikTok.

You know what, John? That is a guy who's probably written 50 articles about how Facebook makes kids have eating disorders. And so here's a Facebook product, gonna cause eating disorder too. You know it will. It's in their business model. All right, I'll go last words and then we're leaving. Bryce— no, John Watt, you already gave your last word.

No, I was doing you. I said Bryce.

No, it's Cody first. We said best for last, so we're giving it to Bryce tonight. You missed that joke, John. I made that joke at the beginning of the last words. Um, it wasn't a joke. Bryce is incredible. So let's go with this. I will go with— I think compute Solves everything. And the takeaway from an investment perspective is that this is what Bryce and I were triangulating earlier today in a discussion, is that the soft— the— you have to be as close to like the actual physical goods to keep value. The further up the chain in just software that you're going, You're doomed. So Duolingo, pure software, no physical goods, doomed. eBay, less doomed. Uber, less doomed than even eBay. All of them though are threatened because the cost of, of making their competitors has gone to zero. And so there could be a company that's got $10 billion, Apple, $100 billion in their checking account, they could tomorrow be like, hey, you know what we ought to do? Let's turn on drivers and make our own Uber app. Like, there you can, you can pay people to get the stuff if your cost of launching an app— the cost of launching satellites is going to minimus. The cost of launching any software has already gone to minimus, and there's application, there's implications, ramifications of that that we're trying to parse out because that's where trillions of dollars are on the line. Bryce, last words.

Yeah, I'll echo that because I think what I'm thinking is that like in, in some ways the software revolution of the last 20-some years, um, like the idea of 90% gross margins where you and you and you build a product, then you can sell it infinite number of times for a price, and your cost— you incur the cost one time, you sell it an infinite number of times at that, basically zero marginal cost. In some ways, it's almost a tax on the real-world economy.

Yes, it's a constraint, right?

And so what we're saying now is there may be Like those, all those companies that were making 90% margins with organizing information and software and things, like what if that gets collapsed? And then like those margins get distributed across the real world of things moving and things happening. And there still are going to be tech companies, technology companies making 90% margins on things. But it's not going to be like if you're making 90% margins on this low value-add type of thing, that just, that just doesn't survive in AI world. And your mark, that, that margin gets redistributed across the real economy, real economy. And I just, because historically there weren't companies with 90% gross margins. So for 99.9% of history, that, that wasn't a thing. And then it was a thing. And it's almost like maybe this is almost like a reversion to the mean type of thing happening.

Does that—

any of that makes sense?

It's first principles of economics and efficiencies, right? The constraint, the co— there's been a tax, a constraint, a charge, excess profits that software companies have made for 20 or 30 years. And it was always a matter of time because anything that has constraint and has outsized margins both their customers and their competitors will spend all their time trying to at least compete and reduce that friction, that constraint, that cost. And that's exactly what AI has done. What's crazy is AI is doing it on a scale that's global and happened like it, instead of it taking 10 years, 20 years for Tesla to compete against GM and Ford, It takes ChatGPT 8 months to get a billion users and takes them 4 years to compete against Duolingo. But we're here. It happens so much faster. That's good deep thoughts. I like it, guys.

Yeah, I'm gonna think about that some more, but I think that's an interesting thought I've never had before.

I had several thoughts I had never had before because of today's discussions. Even though Craig got mad at John earlier.

I get mad at John, felt like it.

I was kidding you. All right, with that, I leave you all with this.