The AI Revolution Call: AI Workflows - How has AI changed your day to day?

You've outdone yourself again. Part of the AI call is the 15-minute— 15-second intro and outro video each week.

That's fantastic.

I like the first one better, but I thought this one would be fun.

Yeah, I agree. The first— the last— the prior one with, uh, he dunking on Michael Jordan was maybe slightly better. This was completely shocking.

I didn't know where he doesn't know what's coming. That's good. It didn't go landscape though. I come— my hand was showing.

Yeah, I don't know why I did that, uh, but whatever. I thought it was very cool though.

That was great. Welcome to the AI call, the AI revolution call, I should say. I'm all about the revolutions. I am Cody Willard, Mr. Revolution Investor. Literally trademarked the term probably 20 years ago and then let it lapse. I don't think anybody's stolen it since, which may— we should look, make a note, Craig. Yeah, I don't even want to say it on air. Everybody will run out and trademark my term. Anyway, yeah, welcome to the AI Call. I'm Cody Willard. Bryce Smith, Greg DeLonn work with me in Revolution Investing at 10,000 Days. We run funds investing in tech like AI. And John Dio is a life— not quite lifelong friend, but 30-year friend of mine who works in the AI industry for decades. Yeah, I said decades. What do you guys want to hit on first? I played with Manus this weekend. I, uh, have been spending a lot of time— Meta bought a company called Manus. Um, it's an AI agentic platform thingy. Picture Gemini on steroids. Uh, you give it a thing to do and it log into your accounts, you can do all kinds of stuff. What I spent the weekend playing with— I'll pull it up and show you guys actually. Ai-discovery.manus.space. I'll share my screen. That is not it. Must type the wrong thing. Okay, I'll pull it up here in a minute. Um, Craig, Bryce, John, have you guys changed how you use AI in the last 6 months? Because I sure have. It is so much more an integral part of every day. There's— we put driftwood in our little betta fish tank that our— my mother-in-law bought for us at a hardware store months ago. And initially it was in a bowl, then it was in a 5-gallon tank, now we have a 10-gallon tank. Put driftwood in it and these little growths came out of the driftwood. Took a picture. 'Gemini, what is this?' 'Oh, it's sugars coming out of the driftwood. You didn't like soak them long enough, and it's not harmful to your betta fish. It'll be gone in 2 or 3 weeks. You could get a snail to eat it, and the snail betta fish.' I was like, 'That's good info. I think we're gonna get a snail for our betta fish.' I say betta. I think most people say beta, but it's spelled B-E-T-T-A. My wife makes fun of me for saying betta.

Yeah.

Yeah, yeah.

When I'm all day, every day, it's just, it's more and more. No.

Yeah, for me, it definitely is that way. I mean, I use Gemini at home, like an assistant, kind of, you know, if I need help with this, that, the other, or just like cooking times, you know, random things on my car, if I'm trying to just fix something real quick. But from day-to-day workflows, it means It's completely changed the way I work. Every time I do something that is about to take me more than 5, 10 minutes, I'm like, am I doing this the easiest way possible? Am I optimizing my time by doing it this way? And is there a workflow that I could better have? And do I have enough Manus credits to make this happen right now?

Right.

I've had to increase my Manus credits several times now.

And I built a, I built a good, a solid structure of a customer resource management tool that, um, we're going to hopefully replace our current one, and then we'll never have to buy or upgrade or anything because we'll just have Manus code in any upgrades that we need. And the Agentic framework is already there. Um, I'm very excited about it. There's a lot that I want to that I want to input into the tool, and Bryce has been giving me some fantastic feedback. So I really do think this is going to be a, a great fleshed-out tool for us when we're talking with investors, keeping track of, um, you know, investment management and where we're at in the operation side of things. So very excited about that, and I wouldn't— would not be able to do that with, with, uh, without AI for sure.

So what do you, um end up spending on, on Manus when you have all this topping it off and stuff during the month? What do you think? How much goes into it in a month?

I'm, I, I'm at $100 a month. I've, you know, I started at $40 and then it was $60 and then $80 and now at $100. And I ran out of credits this weekend, but I didn't go to the— it's annoying, you have to go to the desktop on Manus to buy more credits. If you try to do it on your phone, it's like, ah, that doesn't work. It doesn't even tell you this only works on the desktop. But you would think an AI might actually be able to, like, come up with a better prompt when it tells you you can't buy it. Bryce, John, you guys live in AI.

Yeah, I think, um, I mean, I use AI so much. Manus and Gemini, I guess, are kind of my two primary ones now. Um, and I was thinking the other day, like, at this point, I think if I was the CEO of an enterprise like a basic, at least a $20 a month subscription to your Copilot or Gemini, like, I feel like that's just table stakes, at least for your white-collar employees, maybe not your, you know, guys working in factories and things. But, um, and so I like kind of asked, you know, do some analysis on the, on how many companies have actually implemented it across their professional workforce. We lost John.

He needs Starlink.

But I don't remember the number. I have to go find the— where I was doing that research, but it was like, it has to be maybe like 10% of like white-collar enterprise employees have access to an employer-paid-for chatbot subscription. I mean, Microsoft's Copilot only has about a 3% attach rate in their— amongst enterprise.

And rephrase that, the only 3% of the companies that use Microsoft 360 are using Copilot when they're using—

yeah, or at least are paying for Copilot. I think they have like a free version.

If you have Outlook, you get some version, free version, which is crappy and nobody's using either, right?

And I mean, to me, the ROI on a $20 a month subscription for white collar, like if they can save an hour a month, that pays for that, you know. Um, and so I just, I think there's a lot of low-hanging fruit there. I think companies move slow, enterprises move slow. Um, but when people talk about like token usage and how much compute demand there is, like, like that number is going to be 80, 90% within the next couple years, I think.

I don't know about the next couple years. I mean, the corporations, enterprises are so slow adopting new technologies. But look, I mean, it's only going to grow. I mean, even if you double the use of the number of things, that goes from 10% to 20%, which is huge growth. I mean, that'd be a doubling. So let's call it, you know, if you get to half within 2 to 3 years, it's 5x.

Yeah. And I guess what the shocking thing for me when I saw those numbers was just like we talk about agents and content generation, all this like super advanced stuff, building apps that we're doing now with the AIs that we use. And, and most people, most companies, um, are still not even doing the basic chatbot thing, you know. Um, and so it's like, oh, AI has to get so much better. I'm like, no, I think it's, it's going to keep getting better. And it's just a matter of time as companies figure this out and see the ROI and adopted. And I mean, that, that alone is a lot of demand for compute, not counting agents, not counting Netflix making movies, video games, everything else that's coming down the pike.

Just, just more usage on an enterprise level would be exactly that, right? And it's— it like, I mean, the thing is that there's all still so much more demand than there is supply of compute that OpenAI killed Sora the video generation part of their business because they— it was wasting too much compute power and not generating enough revenue.

Yeah, and I read also it was costing them $500,000 a day at one point.

Wow, that sounds low. That's only $6 million a year— a month, right? So I bet it— what is that, $15 million? No, I bet it's— I bet it was hundreds of millions, Craig.

That's probably just the reported number.

Yeah. Yeah, it is just always amazing to see how, like, AI— Bryce, what I was going to say, what you were just saying is it's AI is getting better faster than enterprises are adopting it.

Yeah, there you go. Yeah.

John, you're muted.

Yeah, I'm trying to share a screen here. Like, that's a good lead-in for— I don't know if this will go.

Let's see.

This is—

we got it.

Is it sharing? This is Anthropix Claude releases in February, March. They released like more than one thing a day. Over the last 2 months. They're going crazy. And so people who are using them on the enterprise side or the technical side, uh, it, you know, they've made it basically, uh, do what OpenClaw does, um, but in a safer way. Um, the leaders on the coding end of things, especially when the coding is happening in teams. So for me, this has been my change, has been using more and more of Claude.

And they're releasing stuff as much on Saturday as they do on Wednesday.

Yeah, I don't know why that's dark on the Saturday. I guess that's the fact that it's a weekend. Yeah, you see, uh, some stuff coming out there.

That's incredible.

Yeah.

So, John, are we at the level where we can even discuss Singularity and the concept of AI can just continually improve from here on out?

Uh, definitely hearing it from not crazy sources. They're talking about it that way just recently because what they're saying is now you can see examples of the AI building the AI, and that's the kind of first step along.

And then you get the AI that builds the AI that builds the AI.

Yeah, right.

I mean, I think Jensen said in an interview that we've reached AGI.

I don't know what it means.

So Jensen has a better understanding of it than I do. I can— that's all.

Well, I put a couple other people maybe at least on the same level with him of understanding, and they don't say that. So, you mean, you know, but, um, what is it, you know, what does it mean? It's a moving target.

Well, Bryce, you're the one that pointed out a year or two years ago at this point that intelligence is on a spectrum, right? And we're certainly moving towards the spectrum towards AGI if we're not there yet.

No, he was saying he's on the spectrum.

Probably of being an artificial general intelligence.

No, just no.

Here's the app I built with Manus over the last few weeks. Let's see here. Present. Share screen. Oh, that's not the screen I meant to share. That's it, but I got a better version here. Oops, now I lost you guys. It's my first time using a computer. Share screen.

We're still here.

We're here.

We're here for you.

Thank you. Thank you. I'll try to figure this out and actually pull it off. AI Discover. There it is. So check this out. I made an AI app discovery. This is sort of my baseline thing when I play with AI coding things. I'm like, make me a TikTok knockoff of that shows me AI apps and games instead of random silly videos. And you can— some of them, they allow the— what the game itself allows it to be in an iframe. So we embedded on the actual site. You can check it out and sign up, watch videos. Some of the game— some of the stuff is actually games. Why won't this go back now?

No.

Huh, there it is. Um, you can go to games, search through the games, click through, tells you all about it. Gives you an idea of the app's DNA, what, what it actually is for. You could— people could leave reviews. Um, we have the ability for people to build stacks and be like, hey, I use this video generator and this music generator, and you guys should check it out. And then everybody can share them. It's got guides. Hey, you want to learn about AI Music. It writes a fresh guide from LLMs every week to update and make sure it's current. Obviously, this is the first time I've clicked on this today, or this week. You can also sign in, have a full admin page. No, no, not another account. And then check out the admin page. Oh, admin. And, like, we've got 230 published. There's 1,000. I don't, I don't have it published if it can't find a screenshot or some sort of a thumbnail for every single one because I want it to look good. I can go through and tell it to try again and find it. And then it just added that thumbnail. So we've had some fun playing with this. And I spent $100 of Manus credits to build—

you did it with Manus?—

working website.

You did it with Manus, huh?

100% Manus. I did it also, I actually played the all weekend with Gemini also, asking it, tasking it to build— not Gemini, AI Studio from Google, right? To build a similar thing. And it was, it had, it was a lot more problematic. It was more cumbersome. I spent probably 3 hours on each over the weekend. And this one was far and away better.

So Manus doesn't really have its own app building stuff, do they? Or I mean, like, they're, they're like driving something else like AI Studio when they go do it.

No. Yeah, it doesn't natively within the Manus app.

Okay.

It's got handles. It'll set up your backend databases, auth. I mean, everything literally.

It never asked for an API. That entire website was built without an API key. Without any external help.

Do you know what it used for any of it? Like what the stack is? Is there a database behind it?

Yeah, I think it's just using Postgres for the database and the LLMs. So it's got— Manus itself runs on Claude and Gwyn, a fine-tuned model of Gwyn. But then, so Manus uses virtual machines, right? So all of this is happening in a sandbox environment. Your virtual machine has APIs natively to several OpenAI models. Anytime if you build like Cody's app has AI built into it, it will use what it thinks is the appropriate OpenAI model. What else are we missing? The authentication is a custom version of OAuth, I think, or however that works from Google.

Interesting.

Yeah, it's an incredible app.

I mean, what's crazy is, you know, again, like, I've built app companies in the past, and this would have taken months, years of like coding from professionals in America and some support from India.

John, I have a question. And Craig and I were talking about this last week, we couldn't resolve it. But there's been like this big shift towards the agentic platforms, agent orchestration platforms, right? And they— that's been a big kind of step change. Can the raw— if the intelligence itself, the LLMs themselves, can they get so good and so smart that all of this just happens within the native LLM? You don't need Manus, you don't need CoWork, you don't need OpenClaw. Like, from an architectural perspective, Ken, like, we had the LLMs, then we built the orchestration on top of it. Could, could the LLMs collapse the stack once again if they got smart enough?

Uh, oh, here's, here's the yes and no, I think. Is that even if it's yes, it's no because it's not that they're sticking it in the LLM, it's that they have their own harness that ships with the LLM. When you go and get, oh, I'm using GPT-5.3, oh, I'm using Opus 4.6, when you use that, you're not just using the LLM, you're using that LLM with its harness and its harness does that agent stuff. So you're not seeing it, but it's there.

But like, co— Cowork does things that regular— if you're just using Claude through the web app, what the browser, you can do so much more on Cowork because that access to your files, plugins, integrations, all of them, not plugins, integrations, connectors and things. Like, is there any fundamental architectural reason why that can't all be built into Cloud.com or ChatGPT.com.

The way in which some of it is getting, uh, built in is where you have the, uh, large action model. And so the more that you have these multimodal things, uh, the— so it's not just a large language model, but there's, uh, the part of the model itself for it to like turn language into action. You know, I say I want to go do this thing, I want to open this window, or make this thing happen. So the action becomes part of what the model itself is able to express that it does. So in that sense, it's getting into the model. But I think, you know, I'm always wrong, but I think harnesses are here to stay. I think we're going to continue to see this, the important role that the stuff around— because remember how we used to talk about like, well, where's the value? You know, all the people with— selling the models, they're getting all the money. And it's like, well, there is room for these, whatever they are, and some of them are coming from the model companies and some aren't, like, you know, like Manus, um, that for some of that money to be going to that wrapper that has all the agents being managed and everything, and it has you interacting with the OS and so forth. So even when some of it kind of works its way into the model, there's room for more of it to intelligence to still show up in that outer piece.

Is the analogy like, you need an operating system? Like, you can't just have— like, the raw compute is only so valuable, you have to have an operating system?

That's the analogy. Yeah, I think so. Look at what, what's his name? Karpathy came out with in the last like, what, 2 weeks ago or something? Andrej Karpathy.

Yeah.

Formerly of Tesla and then OpenAI and now on his own or whatever. But he started to— he introduced this kind of reasoning or learning that happens, like, in that outer part.

You can talk—

you can use it on the models themselves, but you can do it, like, the average hobbyist could go and improve performance by kind of doing this thing that, you know, experiments— and I'm forgetting what it's called, something like research, I don't know— but, um, experiments, sees how it did, checks the result, runs another one, and just— and then you're like, just go do that, go get into a loop of that and come back tomorrow and smarter.

John, you segued into a question I've got, and that is, is when I ask Manus to add a feature or do something on that website, even like if certainly on the initial build, you know, I gave it the prompt and then it's like, all right, I'm on it. And it starts listing all of the projects it's working on, what it's thinking about, where it's researching, what it's building. And it took 45 minutes. Sometimes it takes 20 minutes. Sometimes it takes 2. In my mind, that's a compute constraint is the reason. Could that stuff not— if there were enough compute and I were the only person building an app at that day on all that compute out there, could it be— would it take 4 seconds instead of 45 minutes?

So like, first thing is to recognize that they did it almost like on purpose. They kind of— not to make it— make— not making it longer on purpose, but just they they didn't optimize for time, they optimized for doing as good a job as possible because that's what they were being measured on. That's where they went first. Could they tighten that up? I think so because there's things like get a team of agents that are happening at the same time doing different pieces of it and then come back. There's architectures like that. There's obviously the models themselves getting faster. Different aspects of it, like Bryce was bringing up, where, oh, look, look what the model can do that used to require this other aspect. Overall, the whole thing can get faster for sure. Then there's learning like, oh, I did this before and steps 1, 3, and 5 got the job done, so I'm not going back. I'm not sending the mouse down that other part of the maze again.

This constraint. I mean, in a perfect world with unlimited compute, would it take 4 seconds instead of 45 minutes to do the same job? It is a constraint of compute power that makes it take time.

Yes.

Therefore, I repeat on this call what I told Bryce and Craig last week in the office. Until It takes 4 seconds. Like, it will be— it'll be— the analogy is I'm always using is downloading the Michael Jordan or Anna Kournikova screensaver in 1996 took 5 or 10 minutes, and now it's a silly thing to picture. That's how this will be at some point in the future, in the next few years. I, instead of spending 3 hours going back and forth at that app, that task would have taken 10 minutes. The compute power is the constraint. And they're going to build until this stuff is like dial— like broadband internet versus dial-up internet.

No, it's already in a sense broadband because you started off by showing that, uh, app and saying that it used to— would have taken a year or, you know, or months and months. So it's— if it's gone from months and months down to 3 hours, then to say, oh, this doesn't— it doesn't even count until those 3 hours or 4 minutes, like, it already counts.

4 seconds.

There's other things that— and isn't it on a cornucopia? So there's, there's other things that, um, yeah, uh, there's other things that it can do in 4 seconds already. But the thing that you're talking about, you know, takes that. And here now, I— we mentioned it, it's going to show up. But, um, this is, uh, back from Cody's day. This was kind of what he had in his locker.

Yeah, no, I— this is actually— I, I would have been— this would have been in my 20s, so I didn't have a locker when I was downloading an Anna Kournikova screen. But she was like— probably got up to like number 5 in the world in tennis, and she was great, but she was also borderline supermodel. Yeah, then of course I mean, come on, the guy I dunked on last week in the intro video. That was actually— I will say I had Michael Jordan screensavers because I like the inspiration.

Another piece of news that is sort of related to Cody's question about like the compute constraint. The memory issue has been a problem. If you follow hardware, there's been a big demand for DRAM and, and also high bandwidth memory, HBM. And then last week, Google put out a piece of research about algorithm they call TurboQuant. I don't know if you saw this, John.

Yeah.

But basically, I think it allows them to serve something like 6 times more users with the same amount of fixed memory, which is the constraint on compute right now.

Or it makes your local models— you can, you can use bigger models locally or make them faster and stuff. So like, it's usable by the hobbyist as well.

Yeah. What I thought was interesting about it is it's a very technical paper and everything, you can go read it. But the fundamental change that they, that they kind of figured out was in terms of saving like a piece of information, they formerly were using Cartesian coordinates, so like X, Y, and Z, right? And that's where this piece of information is saved with the 3. So you have to record the location of the piece of information. And they said, look, instead of recording this precise piece of information, you know, there's lots of pieces of information in a model. They said, well, we'll just tell the model, give it an angle like on a circle and say, okay, at 37 degrees or whatever, go this far from that point and you'll find this piece of information.

Instead of saying transformation, it's just, it's a transformation. It's like, with linear algebra, when you just rotate the right, how you decide I'm going to describe it in terms of these two things instead of those things, right?

And that saved so much memory. And it resulted in almost no loss of quality of the information. And I thought it was fascinating as a non-technical person to be like, well, that's simple. That makes sense. Like, I understand, like, Instead of— like, this is the line from the paper that made sense to me. This is comparable to replacing, quote, go 3 blocks east, 4 blocks north, to go 5 blocks total at a 37-degree angle, right? It's like, it's as the crow flies rather than direction by direction, right? And you get to the same endpoint. And I'm like, why? Yeah, gotta be able to fly. But like, that's sort of—

was that a Michael Jordan reference? Of course. Come Fly With Me, greatest VHS tape of all time, but available on Netflix right now.

From— I mean, there's a lot of implications, but from an investing perspective, the memory stocks have been just absolutely on fire for the last 6 months. And we've been asked about them a lot. And one of our main things, Cody has said this over and over again, is like, look, this is a commodity product, or at least a commoditized product. And when it's— when a commoditized product becomes a constraint, like memory has, everyone that— all the buyers of that commodity spend all their time, effort, and money figuring out how to solve that constraint.

And all of the companies selling that product spend all their time and money trying to build more.

Right.

So, you know, you sort of have a cyclical aspect to it that at some point the cycle turns, because might not be now, it might not be in 2 years, but eventually there will be more capacity to build more SIM memory chips, and eventually it'll meet the demand, especially if your customers are spending lots of time, money, and resources figuring out alternatives.

Well, I could say that that invention essentially, uh, increased capacity, but it's not like it had any, uh, effect on, uh, demand in the sense that it's not like, oh, oh, we have— we've got enough now.

It's like, right, no, I mean, that's just it, right? Like, the memory chips are still sold out, but what this probably does is pull forward the point that I just mentioned. The it'll get to oversupply earlier than it otherwise would have. And it's probably certainly not in the next 6 months. It's probably not in the next 18 months. But the reason why memory, why Micron and SanDisk and Western Digital trade for years, decades at a time with single-digit price-to-earnings multiples, extremely cheap relative to other stocks, is because it is a cyclical business. It is a commoditized business. And it goes through years-long cycles of overcapacity along with, like we're going through right now, undercapacity.

Well, older overcapacity was— I could kind of trace it to underinnovation.

Wow.

Great point.

Yes.

And like we went through for a period of time there, a long period of time where it's like, okay, so we've got computers, so we've got PCs. Everybody's been using them all this time. There's not much new about it. And all was coming out. And then all of a sudden there's this new thing that like is so thirsty, and it, it kind of moved, moved things. You moved the innovation.

Yeah, it really was almost a dislocation to the supply curve— to the demand curve.

Yeah, right.

Yeah, it literally shifted the model up. Maybe not an order of magnitude, but double, triple something.

Yeah, a lot.

Yeah.

Do you think anybody has ever traded physical memory? Like when, when the— when there's a supply glut, you literally go buy up tons of like DRAM.

You know what's funny, Bryce? No, because that's another interesting point that I hadn't thought about. Like oil is a pure commodity, but you also— like it doesn't degrade. Like you can store it for years and then use it. And if you bought memory, like RAM chips, 10 years ago and stored them and went to market with them now, yeah, people would like laugh. They'd be like, wait, you want me to use a memory? Like HBM didn't exist 10 years ago and you want me to use— and imagine again, and even the price per, you know, unit of storage has all— it's dropped 95% over that time anyway. So your chip By definition, it's worth 95% less.

Yeah, but now use a different, um, kind of gauge than, than, than the spoilage, because, uh, agriculture has commodities and they can spoil some of them, you know, like, so how come the memory is not a commodity like spoiling agriculture?

Well, the price of corn either, like, there's a time limit to the price, like it's all or nothing. And the memory chips still have value, but it's 95% less later.

Like, so isn't there still a kind of a— don't some agriculture commodities have a kind of going, uh, like a rate at which they go bad kind of thing? Sure. Yeah, yeah. So let's say, I don't know, let's say it's a year, uh, so that— and memory, let's say it's a year because a gig is— a gig is still a gig for a while there, whatever. So So now, so that wasn't enough. The fact that it spoils wasn't enough of a reason for it to not be sold the way Bryce was talking about. What's the other reason? Is it people thinking of brand? Is it that there's like—

No, John, I'm not sure what we're disagreeing. I mean, if you buy something for $100 and a year later it's worth $80 every time, like corn prices can go up and down. Memory prices, they're gonna make memory cheaper.

Oh, there's different—

more—

is that there's a qualitative difference in memory over time? There is no quality difference in corn.

Corn, when you're buying futures, whatever, you're still buying at a certain time for certain corn, right? So it's not just— I'm not just talking about it going down over time, I'm talking about like, okay, they both go down, then why is it that there's no like memory futures.

Okay, well, I would actually almost guarantee you could find someone on Wall Street to sell you memory futures. Interesting, right? I would think so. If you went out right now and were like, I'm, I'm open for buying and selling memory futures, there's hedge funds out everywhere that are like, bit like the old guys, 42, 42, 14, 14, 42, 14, and yelling at each other. It'd be brilliant.

It's not like on the futures market the way, you know, so somebody's doing it somewhere, but it's not like, um, yeah, like, uh, uh, bespoke, not, not a, not a product that all of us— yeah, so there's no forward contract kind of thing.

I think that does speak to the difference, and Cody points this out time to time. Memory is commoditized, but it's not a commodity, right?

That I think is probably the real point, the nuance that, John, that we're— that that's the difference. Corn is oil. Those are pure commodities. Memory chips are differentiated. Each brand is differentiated. There are different things.

Like what?

Like what you can— like, say you go buy some DRAM thing, it has to like fit into a server. If servers change dimensions, it doesn't fit in the slot anymore or something, you know, like, I'd like 1000 terabytes of light sweet DRAM. Yeah, right.

Sold, sold, sold. 4242.

On the retail side of things with RAM, the only way that they differentiate from each other is where the lights are on them. It's all marketing at that point. Because you go buy DDR5, DDR6 RAM, it's, it's really pick your— pick the color you like and what brand has the coolest font because it's all the same crap.

And John, you build computers, Craig builds computers. Have you built a computer, John?

Yeah, yeah, I cut my thumb so bad. I didn't realize that they're so sharp on the inside. You know, the little— the cutout holes, they don't get milled. They're just like razor sharp. So my computer, I ended up naming it Ginsu.

What kind of case did you get?

I don't know. This is ways back.

Okay.

The new— when was the last time you built a computer, John?

Yeah, I think it was back in, uh, yeah, 20— it was in the 19s, probably.

Maybe, uh, in the 1900s.

Yeah. No, you know what, it was in the— it was in the— it was in the 2000s. In the 2000s, the 2000s.

Bryce, I like the 1900s thing. I'm going to start using that. Back in the 1900s was the last time I did that. Yeah, John, I'm with you. I'm not that much younger than you. John's like 68 now.

Yeah, that was harsh.

I think someday my grandkids are going to be like, our granddad was born in the 1900s, you know, like he's old.

Right, absolutely.

I used a Laurel and Hardy reference today when at work with something came up and it was like, oh, that's like, uh, March of the Wooden Soldiers. And people are like, what is that?

So raise your hand if you have any idea what March of the Wooden Soldiers is, or Laurel and Hardy.

Have you heard that?

I do know, I know Laurel and Hardy. Bryce, you know what Laurel and Hardy are? Bryce did not raise his hand for Laurel, but he did know what models Manus uses to make my app work. So I think it— he actually knows what's much more valuable than Laurel and Hardy. Yeah, who's on first?

I try to scan it. I got Postgres and, uh, Laurel and Hardy.

Um, there was another piece of news. Keep talking. There was something else I was going to ask about.

Oh, keep talking? Okay, I can do that. No, so there's this game called Mind the Gap, and it's great because it looks like Monopoly with 4 sides, but it's 4 different, um, generations. It's like Gen X, Boomer, Millennial, Gen Z. And the question you gotta answer along that end as you go— it's really pretty funny.

It's a board game?

Yeah, it's a board game. And as you answer, you're answering trivia questions as you land on spaces.

And there's like, it's— there's no AI generation of the questions or anything?

Not at all.

It's on, it's on, it's a board game. And so when my daughter tries to answer boomer questions, it's the funniest thing. She's like, what the heck are you talking about? Like it was a radio show. She's like, what's radio?

A lot of fun. That sounds like a lot of fun. I do find it ironic that John, Mr. AI, is like playing a new board game.

I'm also creating a board game with AI, but a physical board game.

So that's awesome.

I believe it. Bryce, what was the other news?

I lost it. I'm thinking about other things. I can't— I couldn't— I thought it was— there was some news on Anthropic.

What?

I said your mind was racing too fast.

We know compute power. Something on— was there any news?

Won their first round of the lawsuit.

Uh, yeah, yes, I saw that.

Um, the market is destroying all AI-related stocks right now. I guess we should just hit on that quickly. Um, look, I, I— these are the most sophisticated, well-capitalized technology companies in the history of the planet. Take out the word, the descriptive technology. These are the most well-capitalized, sophisticated companies in the history of the planet. And all of them, songs, Apple, are spending more money this year than they did last year, which was way more than they did the year before. And the only reason they're doing that is because the return on this investment is incredible. They're not doing it for pie in the sky reasons. It is happening here already right now. Craig, while I'm talking, would you pull up We'll finish with the, uh, backlog.

And it's for survival too, right?

I mean, backlog app somewhere, Craig?

Yep, let me pull it up real quick.

They're doing it for survival.

He's less than confident about pulling this up.

He's got it.

What's that?

You look less than confident about being able to pull this up.

I just had to think about where it was, but I remember.

All right, and this is another case in point. How long before Craig hits mute in the Zoom call, or StreamYard thing we're doing here and says to his computer, hey, open the backlog thing I've been working on.

You know, I'm gonna do it now, right?

It'll do it. The point I'm going to make here is about the fact that I keep telling you guys that I do not believe that AI is going to kill jobs. I don't believe robots are going to kill blue-collar jobs. I think there's more work to do There's always more work to do. There's always more demand. People always want more. They want better. They want cleaner. They want bigger. They want more. In the world right now, we just started working on this on a side project again. Was this Manus?

Yes.

And these 6 industries, 6 industries, there are hundreds of industries in the world. 6 industries, $7.3 trillion of backlog. You could build every airplane that you think you could build for the next 5 years and you still wouldn't meet the demand. The other thing that's weird about economies and backlog and things like this, the backlog might actually grow as supply grows because there are plenty of companies and people out there right now not ordering things that are 2 to 5 or 10 years out. For delivery. And if supply came forward, they would place those orders. And even if the time to deliver that $7 trillion or whatever number we're gonna talk about comes down and becomes— you get it quicker, the backlog itself might grow exponentially too. And you probably have more backlog. This is all understated. And so I tell the story, I took Jim home at lunch, but my golden retriever, Jim, my wife says to me all the time, John, she says, Cody, how much petting is enough for Jim? And I said, more. She goes, how much does he need? I said, more. It's always more with Jim. It's always more in the economy. It's always more with consumers. It's always more with companies. There's always more growth. There's always more backlog. There's always more demand. Prices come down, more demand. If time to get this stuff comes down, more demand. So just click around a little bit.

Show—

click on something. Click on— not do the, uh, what'd you click? Commercial aerospace. Look at that. The backlog is $1.3 trillion. It'll take you 7 years to meet it.

You're—

and if you buy more airplanes, it— like, you don't think United would like to replace their fleet more quickly? The reason they're in the backlog is because they're ordering the stuff they need. Everything— AI chips, Intel and AMD have sold out of every chip they can possibly get made. For the next year and a half, 2 years. Nvidia can't make as many chips as they want for the next 5 years. John, why are you— why you got your head down? This is awesome.

I'm thinking about this. I'm thinking I have a thought. When— tell me when I can share my thoughts.

Hit it.

Okay, so this supports what you're saying, um, and I've been seeing recently company, you know, headlines like companies have made a mistake if they let people go, this kind of thing. Um, if you think about as these, um, uh, agent-based AI is getting smarter, companies are starting to think of it like labor. But instead of thinking it like labor that would replace people, think about the decreasing percentage of the labor cost that people are, even without getting rid of them. So then it becomes like, I don't gain as much by getting rid of people My, you know, if I like take the extreme example, say AI can do 100 times what the person can do, whatever. Well, now my labor cost, because it's, it's not like they're free to get them to do that, uh, but, uh, there, there's a labor cost for, for people that's maybe, uh, 1/20th the labor cost for those 100 times AI. So the AI can do 5 times what people do, but, you know, they cost a certain amount, whatever. So now I go, I could get rid of that person, but what do I gain? Not much. The percentage of labor that they take up has decreased.

Well, and what we've— when Craig came on board, we have so much to do. We need another 5 Craigs. Craig is already 2x more capable than the average back ops analyst utility everything guy that he is, because he's using AI to build. Like, I'm like, hey Craig, I want to know backlog in the economy, help me figure this out. And he comes up with this thing, right? Like, this stuff is happening over and over. And we've been making the joke, we quit making it because it's rude, but they're like, well, every time Craig would like build something with the guy, we're like, well, we don't need Craig anymore. Sorry, like, no, like, there's so much more for Craig to do. We need another Craig.

That's the goose and the golden egg. But, uh, it's like, if you've got to though be able to turn your productivity into profit in order to be able to go get the next Craig.

Yeah, it's sort of like the— John, even like in the— we talk about like the 80/20 rule, or like the AI can do 80% of things, but like if you don't have the person to do the other 20%, even if it gets down to like 1%, you can't get the 80%, you can't get the 99%, right? You've got to have somebody that can take it over the finish line. And maybe the finish—

well, Bryce, I love that phrasing. That's really interesting. You're exactly right. Like, just because AI can do 99% of the work, it's by definition it doesn't get the job done, right?

And maybe in the end, that final 1% is just prompting, you know, it's just having the idea, thinking about what needs to be done.

And communicating it clearly to the AI.

Yeah.

Craig, will you share the link with everybody, for everybody with the backlog economy thing? Guys, it's a work in progress. It'll be better and changing, but we'll put it here in the chat. I, I shared the link if you wanted to check out the AI Discovery app that Manus and I built together over the weekend. Um, last words.

I've got a quick video for you on the more thing. It's kind of funny.

Yeah, that's it.

You guys may have seen this before.

This is good TV now.

Yeah. Do you want from me more? How much more successful do you want me to be?

More successful.

How many records can my records break?

More records.

But I'm the best.

But are you a different animal and the same beast?

What the fuck does that mean, Cody Willard? Funny video.

I started laughing when he said it because I was like, I don't know what the heck that means.

It's not a real video, but somebody put those things together. It's pretty funny.

You know, did you guys get the monk video last week?

Oh, yes. Yes, I got it. Yeah, I didn't realize it was in another language. I think it was— was it in German?

Yeah, no, it's, it's, it's something like Norwegian or something.

But what are we talking about? It—

we shared it around. He didn't— he get it. So this is about the— it's a video showing this, like, monk, uh, when books first came out, when until then—

oh, oh, oh, yes, okay, they did everything with scrolls. Yes.

And then this guy's like, he turns the page and he goes, wait, we just lost all those words. It's like, no, no, they're still there. It's, it's a technology thing.

Yeah, it's memory. Turns out we've had high-bandwidth memory since the printing press.

Yeah, yeah, that's right.

All right, that's my deep thought, final word. Craig?

Sure. Today I read an article about Eli Lilly really spearheading drug research with some other new native AI companies. And I really think this is kind of the first foray into healthcare, just a true healthcare for all option, you know, once Once cost of healthcare, we can, we can look into drugs, get drugs to trial quicker, and then we increase access to healthcare through robotics and AI. I mean, I think this is the first inception of that, and then you'll have more people with access to newer, better drugs. I mean, just in my, my personal experience, I've battled with some skin issues my whole life, and I've had to wait years for stuff to get the trial, and just now, I mean, it's been like 10 years I've been dealing with this stuff, I finally found a medicine. If we could have shortened that up to a year, year and a half, this would have been just like a blip in my story. So I'm looking forward to the future where other people may have just annoying or terrible illnesses or things that are just chronic that keep them from living, being themselves, and then that's just gonna be a thing of the past we're gonna be able to get drugs quicker and disperse them everywhere to everybody at a much lower cost.

Craig, it's a great point. I mean, it's going back to my, hey, we gotta fire all the doctors because they're 90%, they're 10 times more productive. No, how many more people need to see the doctor than ever get to? How long is the backlog to go to such and such specialist and everything else? That stuff that we need more doctors who know more stuff and can learn it quick, more quickly and distribute it more quickly and do the procedures more quickly and get you better drugs and better treatments, more effective. Like, it's insane.

You—

anybody looking around the world today and seeing people suffer and it's like, well, AI is probably going to kill all the jobs. I'm not sure. How, what, what are we talking about?

Right.

You know what, I'll agree that AI and robotics are going to kill jobs when nobody is suffering.

There you go.

Let me know when that happens. Bryce?

Yeah, I don't know. I guess I'll just— I think there's some incredible opportunities in the market. I think, like I told Cody today, that the market has been holding this thought in its head that AI is not producing high enough return on investment. Therefore, you should— the hyperscalers and the big AI companies are screwing up by investing so much money. And also that AI is so powerful that it will kill basically every software company. And instead of that happening for a blip, a moment in time or something like that, those two thoughts have persisted in the market for like 6 months, maybe a year now at this point. It's been creeping up and that's gonna resolve itself one way or the other. And I obviously think we're on the side that says AI is producing incredible ROI and a lot of software companies probably are at risk. They're probably not all screwed, but they are at risk, right? And so I think, that is going to resolve itself. We'll look back in 5 years and be like, wow, that was really stupid when everybody was like, why are the AI companies investing so much money?

One or the other. One or the other. It's either going to be like, wow, they were really underpriced, or it's going to be like, well, it turns out AI was stupid. Yeah, you guys laugh.

In which case, the software companies would rip, like, you know what I mean?

That's right, Johnny.

Well, also last week on the medical front, and I didn't look too deeply into this, but there's a guy who had Lou Gehrig's disease and was losing his ability to speak and was getting very frustrating to try to even just talk with his wife or anything, kind of mumbling. And Neuralink got him speaking, or got, you know, figure out like what is he saying So it kind of enabled him to speak. So there's more, continuing on that line of prosthesis of AI. And the end story of the prosthesis thing would be, Kurzweil talks about, which is, if we do get to AGI and ASI and computers get so far ahead of people, one thing would be for people to try to keep up. By being enhanced in various ways.

Yeah.

So that's a possibility.

Yeah.

We saw that article today about exoskeletons and different things too, you know, enhancing physical capabilities and health and yeah, memory.

You know, Bryce, take that thought. I was going to say the same thing, Craig, that Bryce sent out an article to Craig and I this morning. Maybe if you'd grab the article and share it in the chat also. And it's about Ukrainian soldiers using off-the-shelf consumer exoskeletons that have AI in them. They're motorized and they give you— can run up to 12 miles per hour wearing it and you can, you know, carry It's not gonna help with the deadlift and the arms part, but you can carry things faster and easier with this supplemented walking with the exoskeleton you're wearing. I immediately saw that. Bryce found where you could purchase it. You can buy the exoskeleton for $800 on up to $2,000 for longer battery and more power and more AI settings. And we need help. I, you guys have heard me talk a lot about Ameris, my 10-year-old medically fragile daughter, is heavy. She's 55 pounds. I'm 54 years old. My wife is 50. And we— it's hard to carry a 55-pound kid at all, you know, every time you need to move her off the bed or carry her to the chair or, you know, walk, put her in her walker or whatever we're doing. And I've been talking about robots and I need an Optimus robot. And when I saw that exoskeleton thing, I had the epiphany today. Oh my gosh, that would actually at least help us get her in and out of the car where you need to carry across rooms and things. That's not really what the constraint at our house is. The constraint is the deadlift and picking her up with your arms and whatnot. And so then I actually went to Manus. And asked it first, will this exoskeleton that I'm looking at help with carrying her? And it said no. And so then I said, well, what would— are there any exoskeletons that would? There, there are actually there, but you know, I strap on your upper body, um, but the $10,000 for a good one. But you know, again, we need more. We need exoskeletons. Do you know how much more How much better my life would be if there were a $500 exoskeleton that actually had both the legs and the upper body stuff. And I don't even need an Optimus robot, I just need an exoskeleton.

More.

I thought I had seen something where— I thought I had seen something where the weight is, um, kind of displaced by the, the hips of the bottom. It's like there's something coming off of that part of the bottom half of it. That is, you know, catching the weight.

So when you're carrying her already, yes. But I, again, my, my issue is my lower back picking her in and up out of a bed or in a chair. And so it's— but look, in the grand scheme of time, somewhere in the next 5 to 10 years, it's inevitable that I will be able to buy a $500 exoskeleton motorized AI thing that I strap on and helps me when I need to move Ameris.

Think of this, when, uh, the thing you just did with that application you made was bits. The thing you need is atoms. As atoms catch up to bits in terms of people being able to do what— make what they think, then you're going to have what you need.

And let's rephrase that, hardware versus software, just to clarify for the people that didn't— weren't sure what you meant by atoms. We're talking hardware hardware, Bits, you were talking software. And yes, it's actually one of the things that I think the market figured out somewhere in the last 2 to 3 years ago is that especially in technology, there was a dearth of supply for hardware innovation. And there was probably a glut of software innovation.

A plethora.

Maybe not innovate. There's never a glut of innovation. A glut of software product. And I think that's, you know, one of Elon's geniuses was recognizing that vertically integrating and creating an American factory and fabrication and physical products— Adam's company— was a unique opportunity.

Yeah, I think it's harder.

Yeah, I agree. I mean, especially at this moment, I can do that in 3 hours with an app with no coding, with a Manus App with no coding experience. Certainly there's not a place where I can go 3D print an exoskeleton with AI right now. I would bet within the next 5 to 10, 15 years, I can custom order my own personal AI-assisted design for the exoskeleton and have it delivered by drone within a day. All right, that might be too quick. Give it 3 days, couple days for making it.

4 seconds.

4. And with that, everybody,