The AI Revolution Call: Why is AI Tanking the Market?

What'd you think, Johnny?

Sick.

That's pretty cool. That's pretty cool.

All I think is that it's then followed up by just seeing us, so we got to work on that aspect. I got to get my green.

I put a suit on this week.

Nice.

Yeah, yeah, and we, uh, I had some meetings this morning, so.

Okay.

But actually, now that I did it, I think from now on, guys, suit and tie required Mondays for the 3:00, 5:00 PM Eastern call.

Got it.

I don't believe you, John.

I don't think you'll do it. Um, show up that way and you guys don't, then that would be bad. So that would be—

but no, I might start doing this. I feel good. I feel like I'm back on TV. Um, I am Cody Willard. This is the AI Revolution call. Bryce Smith, Craig Gallon, they work with me. That's why they're in the same room. John de Oliveira, 30, been friends with my, with my boy John for 30 years, but, uh, he lives across the country these days, so we're not in the same room.

Yeah, 30-something. And you know what, I can, you know, it's 30-something because I'm coming up this year on my, uh, 30th anniversary, and you were at my wedding, so, in my wedding, so So yeah, that's, I think, probably 32, 33 years.

I know John, and he's been working in AI for 30 years. I was with John one day when he was like, dude, you know this AI stuff? And I'm like, artificial intelligence, what? And he was like, yeah, man, it's real, it's coming. And I was like, okay, let me know when. And it's— oh, that's a good dog.

I just don't like him showing up on TV with his eyes up. No, on— he's fine on TV, but not with his ear flipped up.

He's gotta—

there's a thing I have. He's got to be, uh, properly dressed. He's got to have on his suit and tie too. So yes, um, oh no, you know what, he sees dogs on TV and it makes him like jump around.

Oh, well then Jim will stay hidden. Yeah, um, he was staying hidden anyway. Listen to me. All right, um, That's a good dog. Yeah, that makes my heart happy seeing him back there. All right, guys, I think there's one topic to hit first and foremost and pretty much to dominate this entire discussion today. And that is, what industry is AI not dooming? Because apparently it's doomed everything on Wall Street. It's doomed every AI, white collar, I mean, every white collar job. AI is going to kill everything. There will be no work left for human beings. There's nothing to do. You don't have to work. I don't have to work. Universal basic income. Money is meaningless. I buy none of that. John, what the hell?

You have one industry I know of. AI. Yeah.

AI will not kill AI.

Right. I mean, it was the, the so-called bubble, and if it's killing everything else, how can it be, uh, a bubble? Certainly has bubble aspects, but, uh, it could, it could take everything down and then itself.

But well, I guess it could, right? Like, I mean, we talk about we want to own the oligopoly that's going to be the platform for all AI, and that would be Meta, Google, Amazon. Anthropic, NVIDIA. Then, but NVIDIA is not owning the cloud. Like, the guys who everybody is actually going to be building their cloud, the cloud upon which it's going to be built, are like those three.

Microsoft.

Microsoft.

And the name—

I know you didn't want to—

I didn't want to mention Microsoft.

It doesn't have to do with the ones that are actually providing the technology.

I just wanted you guys to validate my feelings about Microsoft and say no, it's not one The point is, whether it's 3 or 4, it's an oligopoly. There's only going to be a few.

Why is it just the clouds? What's that? Why is it just the clouds as opposed to the companies that are providing the AI technology that's getting it built?

NVIDIA, first and foremost, is the platform upon— you know what, let's just go through the steps. Perhaps Taiwan Semiconductor is the first platform upon which all of this is built. It's a monopoly.

Or that Dutch company, ASMR, what's it?

The ASML.

ASML.

All right, fine. We'll start with the ASML equipment that gets installed at Taiwan Semiconductor.

What about the gas that comes from Tokyo Electron and those machines? Yeah, you're right.

Of course, Bryce.

And what about the miners who mine the gas? And what about the guys who supply the shovels to the miners who mine the gas? What about the guys who supply the wood to the guys who supply the shovels?

The point of the story is economies are magic.

And if the—

so wherever you want to start, let's start at Taiwan Semiconductor because we need somewhere arbitrary, if nothing else, to start. Taiwan Semiconductor makes the chips, and then NVIDIA sells the chips, and then the oligopoly that I'm referencing builds data centers that use the chips. And the only companies, the only entities on this planet that have hundreds of billions of dollars to spend making AI infrastructure work are the cloud guys. That's why I was talking about them.

Well, I wonder, no, I was asking about companies like, I don't want to say OpenAI, but Anthropic, the ones making—

Those guys are— there will be more competition at the LLM agentic layer than at the actual infrastructure layer, because anyone can start an agent company. Anyone can start an LLM. You might make a better one if you have more capital, but you could still make one with limited capital. You cannot build AI infrastructure that will process trillions of things every second and make AI videos and robots work. Unless you have hundreds of billions of dollars to spend.

I thought everybody had hundreds of billions of dollars.

That's in the future, apparently.

Oh yeah, or they don't need it anymore, right, in the future?

Apparently.

So, I don't—

let's start— I don't want to get too off track by talking about why I have such disdain for the philosophy of universal basic income even being possible, much less utopian. But we're nowhere near that right now for sure. And what is happening for sure, and what I want to actually hit on, I think why people would want to watch us, apparently we get no viewers if we're not on X. What is this, 4 people watching us?

Yeah, usually all of the viewers come from X, like hundreds.

We're not on X right now.

Yeah, apparently we're not because there's something— there's a glitch. No, who needs software developers anymore? Apparently X needs software developers because not only StreamYard, but most third-party apps that plug into Twitter X video are not working anymore.

Well, look, yeah, I'll catch you guys later.

Okay.

I didn't realize. I didn't know.

No time for this unless we're on X.

Nobody ain't got no time for that.

At any rate, look, the point is that this is actually a good example of why there would— like, we still need software developers. No software developer who wants to work is losing his job right now. Maybe it's not at the same company he wanted to work at, but there is more demand for software developers than there's ever been, and there always will be from here.

Always will be?

Well, maybe not always. For 20 years from now, henceforth.

No, and give just a little bit of context about what Cody's talking about though, because please do that. The— in the last 2 weeks, and it's accelerated, um, just rolling micro crashes and crashes in the market, um, about which sector or sectors AI is going to disrupt next. Um, and I'm not kidding, you know, it started with SaaS, software as a service, uh, and then we had like real estate brokerage firms and then trucking brokerage firms, logistics firms And then what else did we have? Now, today was banks, cybersecurity. Cybersecurity was last week.

Pretty much anything when there's a press release from Anthropic or anyone else saying, hey, we've got some technology that can do some cybersecurity stuff. People take every stock in the sector out and shoot it, which in some cases is probably right. There will be some software as a service that's never coming back.

Yeah, but like today it was American Express and Bank of America, Citigroup. Like these companies, like it's just indiscriminate selling. It's algorithmic trading probably. And it's fearful, fear-driven trading. Nobody is honestly like, oh, Citigroup is a sell and, and I don't know, Walmart's a buy or something like that. Maybe they are, but it's can't be the majority of people. Most of the selling is just— and just like the sellings get started because the algos get triggered by whatever Anthropic or anybody else says. Uh, oh, today was consultants too.

Oh yeah, consultant slam.

13% down on IBM, historic move. And, and then that's the other side of this. These aren't like your normal volatility. Like IBM is down 33, 35% in 2 weeks.

2 weeks.

And that's a stable— like this is the definition of the blue chip. Blue big blue, blue chip.

I would—

I don't know when the blue chip term came around, probably long before big blue. But anyways, uh, they're like the— the logistics companies last week. C.H. Robinson, an old trucking brokerage firm, was down 25% at one point intraday. It was its largest crash ever.

What's, what's crazy, but about all of this is how, like, if you've been watching our AI Revolution program for the last 2 or 3 years, we've talked a lot about how there's a lot of like unsophisticated software applications out there that are going to get completely demolished by AI coding. It took 2 years before the market suddenly somewhere in the last 6 months was like, hey, you know what? There really is a lot of uncertainty about a lot of what is traditional software as a service, and we better panic about it. And The worst part of our job as hedge fund managers is we're supposed to short stocks that will go down. And we were predicting that things like even Chegg or Duolingo, Adobe, we've been short some of these names, IBM off and on for years. And the market finally caught up with us since 6 months ago. And now they've been unrelentingly selling them. Um, but you know, we didn't make as much money on the short side as we should have.

Yeah.

If any, a lot more.

Yeah. Yeah. It's tough to be short for a long term, right?

Yes.

Not necessarily, Johnny. There is, I'm still short of stock, uh, used to be called Valiant. I don't even remember what they call themselves now. I was short at $300. It's now at $6. I think it's even done us, reverse split at reverse stock split at some point. Point being, I have like $12 of exposure in that name at this point, and I don't plan on ever bothering to cover it. But there is a mistake when you like, for example, I was short Carvana from $100 down to $3, and then I was like, you know what, I'm never covering this.

And it went back to $500. That was a bad not cover.

And then what do you have to do?

Well, I ended up covering it at like 50 instead of at 3 because I didn't want to continue losing money on the way back up. And then I shorted it at 120 and covered it at 140, and then I shorted it at 250 and covered it for the final time at like 280. And I don't have a position in it anymore. I still think it's actually a zero.

They're Let's just say I don't trust the numbers over there.

But John, like, the, the AI is sort of just creating mass uncertainty. And nobody knows. Okay, how to deal with it.

AI's fault.

No, like it. Well, it's humans.

Well, the markets hate uncertainty is his point. The AI is causing uncertainty. About a lot of companies and a lot of industries. And so the market is like, wait, these things have been priced for perfection and/or at least very good futures. And now those futures are less certain. We better take the valuations down.

It's funny, but it's the way we talk. We say AI is causing uncertainty when in fact people are uncertain about AI.

Yeah, that's a good point.

I mean, the AI didn't do anything. To cause uncertainty, did it? What did it do?

It coded some stuff for us.

Like all it did was grow.

And, you know, maybe one of the things that I've been talking about today on several calls, thus the suit and tie, is that, you know, the entire premise of AI is simply making it easier to use, making all services easier to use, making computers easier to use, eventually making cars and robots and everything in the world easier to use. And like, that's the miraculous part of all of this is yes, like somehow whether you're using an application like Salesforce CRM or whether you're making your own software that does your CRM, the point is that you're gonna talk to it verbally, maybe see some graphs if you wanna look at them. But you have the AI just making your life easier.

And I think it's demonstrating that it's definitely— if it didn't, you know, kind of fulfill every promise or something, it is heading in the direction that shows that it will make things easier and starting to make things easier, etc. So that's actually happening. And that is a detriment to all of these companies that spent all this money over at least the last decade, maybe decade and a half, the SaaS world, saying that we're gonna make things incrementally better and we'll do so by spending these billions on manual software development.

I guess what I envision is those companies could and should and probably are sitting around being like, hey, let's use AI to race ahead and make these products even more full of great features and applications and services and agents. And they should not just be sitting there and be like, well, crap, now our competitors can do what we spent 20 years doing in 20 days, so we're screwed. No, spend the next 20 days making your stuff where it should be going anyway. You would think those companies that are 20 years old of selling software as a service would know exactly everything their customers might possibly want to integrate.

Yeah. So you just said it though. They're 20 years old. So true, they know everything, but as a giant oil tanker, can they turn on a dime?

That's what we're talking about with CRM. Like CRM's got 80,000 people. Like it's very hard for them to turn all of those people and get them going all in on AI all at once.

The, the, the, the value of like a Salesforce is in their 20 years of experience dealing with every edge case that comes along. Now you don't have every edge case and it's no longer your concern to handle every edge case. Really, the, the value prop is shifting. You know, instead I want my CRM to do the basics but to communicate effortlessly effortlessly with the other apps and to, um, you know, deal with things that, that come up. But let me talk to it to make it happen and so forth. So it's like, yeah, it has all that knowledge of all those edge cases, and I'm paying for every edge case when I just have these ones over here. Somebody comes along, can make the, the one that's more focused on— you know, I can get my bakery CRM that's literally just a bakery CRM because a baker went and made it last night. You know, it's a little different than I'm going to go buy Salesforce, which over the years has become good at bakeries and everything else.

John, how hard is it going to be for the random bakery CRM application creator to maintain his app and make sure that when Java gets updated in the cloud, that his app still integrates seamlessly with—

Yeah, well, I mean, less and less hard over time with a smaller and smaller staff. You talk about you know, programs aren't going to go away. I, um, I guess I agree they won't entirely go away, but there definitely will be a different, a changing need and a changing set of people who can fulfill that need. And you go from the people who were, you know, uh, engineer, uh, product manager, designer, and instead you have this one person called builder who can do a little of all those things and can, uh, in talking with the AI, get it to produce what they need, and the— it still is running in a cloud, it still is, uh, even it still is SaaS, you know, but how it gets put there, what gets done to make it happen, it just becomes a much easier, uh, thing.

Like, John, so the agent thing has been so big, right? And I've been working on that for the last 3 weeks. And the way this works, like, there's, there's several out there, you can take your choice. But I've been using Manus, I think it's really revolutionary. I don't know if you've been—

I agree.

Manus.

Yeah.

So, like, you, when you go in there, you connect your tools, they call them connectors, but you do things like Gmail, Slack, Calendar, there's probably 6 dozen different apps you can connect and things like that. So, like, Right now I totally get that and why you do it and it's really powerful. Your AI agent can go in and do things and actually take action in all those apps. But like Salesforce, for example, you have all this data saved in there. You give the agent access to Salesforce and you're like, okay, you know, when I, when these people email me or when I email my leads, save it to Salesforce, maybe tell me who I should email. But then pretty soon that Salesforce is just a database And if Salesforce is a database and you're paying for seats on Salesforce, at some point you tell your agent, you know what, go extract everything out of Salesforce that I've saved over the last 20 years, save it in our own database that you build, and let's roll from there, and I'll cancel my Salesforce subscription.

Yeah, go back to fundamentals.

Like, everyone in their minds is running through that scenario with everything, right?

I go back to the, the fundamentals on it, and you're like, what's my workflow You know what actually matters for my pipeline? I've got to convert these people. I've got to kind of get them through the funnel. And that's the stack changes. It's not necessarily what it was.

Yeah, that's what the market's trying to figure out. And nobody knows for sure.

We're customers of Salesforce, right?

Yes.

What are the odds 2 years from now you guys would still want us to be on CRM? We're not, we haven't been legacy customers for 20 years. We've been several months. We've been signed up on Salesforce CRM. What are the odds 2 years from now we're still signed up on Salesforce CRM?

I think slim.

Yeah, pretty much zero.

And if we had done 2 years, if we rewound 2 years ago and I asked that question, your answer would have been, yeah, well, like it works great. We'll be there forever.

Yeah, yeah, it's true.

So really under— like, we are an anecdote that underscores the entire premise of why people suddenly doubt the future of software as a service, whether it's a value-add. Like, we've tried to be bullish on CRM and owned it until a few weeks ago, in large part because we were like, nah, but CRM, like, it's so— like, they're all in on agents and they've got all of this data and that I really don't need any of that anymore. Manus_, tell them about your CRM and Manus experience last week, Bryce.

Sure. So as I mentioned, I've been on this train for a little while, um, and I like— this was happening every day. Agents destroyed this industry, they destroyed that industry. So I was like, how, how do you build an agent? I've never, never done that. I've used chatbots a lot. So I went down the road and I, you know, we had played with OpenClaw, um, we had played with Claude Cowork, I tried those and I was like, for me, it doesn't make that much sense because those are local software tools and almost all of the work I do is cloud-based. So I use Google Drive, Gmail, all of our software basically is in the cloud. So I was like, there's got to be, you know, I tried a lot of different tools. I tried Zapier, setting up automations in Zapier, Google. I tried to do it there. Probably half dozen different tools I use to try to like figure out how you make these agents. And I got to do some simple things. But then I came across Manas, which MetaBot, at the end of last year. And it was the thing that I knew should exist, like in the agent world, it was, it's just like using ChatGPT, the interface looks just like ChatGPT. But you hook up all these tools. And then the beautiful thing is that it like, it will work for 30, 45 minutes, an hour, and it will set up all these sub-agents and it will open browser tabs. It uses, um, virtual machines in the cloud, so it'll literally spin up a virtual machine, fire up a browser within the VM, and go do work there. Simultaneously it's doing work in Gmail, simultaneously it's doing work in Drive. And I was like, this is incredible. I finally see how the agent thing starts to work. You know, I built— it built amazing websites for me, built apps for me. I have an app on my phone now that literally— an Android app. And I've tried to do like— I've had like 3 or 4 litmus tests that I've tried to run through the AIs over the last year or so. One of them was build this specific fitness app I wanted to build on my phone. One of them was to build a dashboard for the hedge fund. And Manus did all of that. It was no—

in one day. Yeah, like in less than a day.

Yeah, I didn't have to go get API keys. I didn't have to copy and paste code from another chatbot. I didn't have to— like, I'm trying to think of some of the other things we have to do, like on Replit. Like, lots of times we just have errors and the code wouldn't work. So you're like, hey, please fix, please fix, please fix. Manus, I never did that with any of it. And it's powered by the same underlying models, right? But they've developed this orchestration layer, which is really powerful.

That's Meta.

Meta acquired Meta, owns it. Yes.

Yeah. So there's a lot of talk about how jobs now, you're gonna be judged by sort of how many plates you can keep spinning, where the plates are agents, you know? And so if these agents, you know, right now it runs an hour on its own, then in a few months it'll run 4 hours on its own. By the end of this year, it'll run a day on its own. When it's running a day or several days on its own, How's that— how, how different is that from having an employee? And how different are you from being a manager of that? And how many employees can you manage?

That's precisely correct.

Yeah, your white-collar employees are definitely going to get kicked up the ladder with their expectations for skill sets. Like, they have to be managerial, they have to be able to manage deadlines. And anytime your boss is like, all right, what are your agents doing? What is your team doing? You know, you're going to have to answer at any point where they are with their project.

I'll go back to the point I was making earlier that everybody's worried about like all these jobs being wiped out by AI, but that's my— this underscores what I'm trying to get at, that if you can communicate effectively, you can prompt effectively, you've got a big future ahead of you because you can make the AI do a bunch of very valuable, productive, efficient stuff for you. Which theoretically, if you're working for someone, that should be helping your boss and helping your boss's boss and helping the company itself be more productive, more effective. And so, but you don't have to go train on how C++ code works. You just have to go train on how to communicate well so that the AI can use C++ code to do stuff you want.

Yeah, I think Bryce's future is bright. You know, actually, you know, you mentioned like C++, etc.

Um, wait a minute, what's that? Was that like an insult to me?

No, to all of us except him. I mean, it was just a compliment. Just let me compliment Bryce without—

uh, wow, Craig, apparently John doesn't like you.

No, of course I, I like him. So what I was gonna say is, you know, you mentioned C++. So this infrastructure code layer that now the agents are being the ones writing the code, and because of that, like, I was just seeing this thing about how much of this— it happened at Tesla, you know, this code in the middle that was Java or C++ is getting replaced with Rust. Now, Rust was known to be an excellent language that's really hard for humans. So it's like it, it works well, it does its job. If they— if people were to rewrite the Unix or the Linux operating system today, they'd write it in Rust instead of C because how well it does in security and memory management and all this kind of stuff. Well, now, since the computer is the one writing it, it can go and write that. But actually, the person who was talking about this with regard to Tesla said the reason it ended up— you have like more Rust code and less, you know, Java, etc.— is because the code that went away was really the code that— it didn't get replaced by Rust, it got replaced by the LLM. So not even the LLM, but well, the, the really in this case just, just the AI model. So the thing like with the car running totally on a model— this, so this wasn't the car, this was, this was, uh, this was, uh, Twitter. Like X. So the, the, all the algorithms that were, you know, promote this or do this with the, with this incoming tweet, you know, those were handwritten algorithms. The handwritten is gone. Everything is this, is the, like, the mod, the machine model itself, the AI directly determining that stuff. And the code that was handwritten, gone.

And yeah, John, isn't there, I mean, how many layers of the software stack are sort of like enabling humans to interface with the other layers of the software stack. And if you weren't needing human, the human element involved in that, would it be different languages throughout?

I mostly, I think of that as a front-end, back-end thing. And the front-end or UI, you got, if you have humans involved, that's, I'm not even talking about the JavaScript, TypeScript, you know, React kind of end of things that you've got that, and computers are starting to write that stuff as well. But if you didn't have it, that goes away as well. But that's not even— so this is just the backend, and the backend, which was Java, C++, Python, is now 85% Rust, 15% Python. Wow.

John, let's run— I'm putting you on the spot. You had no, no forward knowledge of this question at all, but I'd be curious on just your impression on these companies. I'm just gonna name the company. On a scale of 1 to 10, how screwed by AI is it? Adobe.

8.

Unity.

Interesting. Um, 7.

Salesforce?

9.

Zscaler?

Don't know.

CrowdStrike?

5. Because they can themselves jump in and like, you know, on the side, like you were talking about, like use AI.

Yes.

Yeah. Yes.

Workday?

9.

Have you heard of monday.com?

Sure.

8.

Just, it's more, it has to do with the user base, you know, like, what do they like? That's like a Notion-type user base.

Yeah, app-loving.

God, it's again, it's about like, do they just adopt it well enough themselves to throw the same shell around it to do what they already do? You know, so I don't know.

All right, how about IBM/Accenture/Consultants, which were slammed all day?

5.

Interesting.

They always pop back, you know.

Well, it's funny today with the consultants is I think last night or this morning, OpenAI had a press release that they were being— they were doing a deal with Accenture and a couple other consultants to help get OpenAI into the enterprise. And then later in the day, Anthropic's like, yeah, Codex— I mean, Claude Code can do some consulting. All he said was that Cloud Code can now do whatever I told you. What's the name of that software? The old IBM software? CO—

COBOL?

COBOL.

Yeah, I didn't even know what that was. Apparently, the COBOL—

COBOL—

Cloud Code can convert that into a modern language. That's all. So COBOL is 13%.

So I— yeah, I mean, COBOL has been like the language that much of big business and, and then all of government like ran on for so long it was ridiculous, to the point where it was like they had no nothing to back it up. If something happens, the IRS is gonna go down because, you know, some little sprocket broke in a computer and nobody knows how to do anything with it. You know, that, that kind of thing. COBOL is the, the first one of the first two computer languages.

Oh wow, ever did predate Fortran.

Yeah, I, um, I don't know. Uh, I think so. It was— COBOL was created by, uh, Admiral Grace Hopper, who was— she invented the first, uh, compiler that worked on two different computers. Um, so that, you know, got it to where programs could operate and you could write it and it would— you're not writing it for that piece of hardware. So like, oh, it was a big thing back in the day. We're talking 1950. So that— but that's how long the stuff stuck around. And it was so hard to change anything. And I mean, getting scared, like, what are we going to do? People are retiring and nobody even knows how to do this thing anymore.

Except Anthropic came up with a solution today. Turns out that worry was alleviated by Anthropic solutions. So sell Accenture and IBM.

I'm sure they can do it. Yeah. And all this time, no one could. They could run the COBOL thing like they could emulate it the way that these, um, the vibe-coded things are coming out and pretending to be a desktop of a computer. And, you know, I can click and the notepad works, I can click and the calculator works, and it all came out of the thing in one shot. That's what they could do now with how basic COBOL was.

All right, name one more software company or any company that you want to ask John about.

You know, I think we've talked about Snowflake.

Snowflake, you know, I, I thought they have a better shot than others because, um, and actually, what was— is it 1 to 10 that it hurts them? I'll go— can I go -1 that it actually helps them?

Right, that was what we were looking for.

That's our bet.

That's exactly what we've been doing.

Because I think that they're— yeah, I mean, they're at the end of it where they can get involved in models getting refined and created and so forth. You know how the same way that Anthropic turned to IBM or whoever did and said, help us get— no, OpenAI did and said, help us get into the enterprise.

Yeah.

You know, they could turn to them and say, help us with our, like, you know, pipeline where we have massive data we have to deal with getting from here to there. And like, we're so advanced in everything, but just even though we're so advanced, we need, uh, you know, NVIDIA for our chip. We need these other folks to help with manage the, the enormous data, manage all of the, um, libraries of books that we steal from the whole world in order to go make our AI. They got to put that somewhere.

It made me think of something that's just sort of funny. I don't know that it— you can read too much into it. For investing, but like OpenAI, they use apps and software. Like, they are the largest customer of Datadog, which is an app observability platform. You may be familiar with it.

Yeah, I use it. I don't think it's even that great. I'm sorry, I don't want to pick on them. I'm kidding.

And their whole company runs on Slack. They're like huge Slack users.

No, but it's like they're not writing that stuff, right? So it's normal for them to go and, you know, and just, and just use something like that. But I think Databricks is, uh, yeah, they did their $160 billion.

They just said private round at $160 billion versus Snow at like $70.

Yeah, or today, after today, $60.

Yeah.

Um, oh, uh, you were about to ask a question about companies. Did you have a company you wanted to ask him about?

I was saying Snow, but he said Snowflake, so we're good.

Um, what was I gonna say?

Um, I've lost the thread. Johnny, like you've not heard of Zscaler, but you use—

Datadog.

Datadog. Do you use Snowflake? Do you use Databricks?

Snowflake, previous life, you know, and I loved what they were doing with what I want to call it, I think it's called a snowflake schema, which is a kind of an advanced form of a star schema, which is a way that you store your data. And it let them do awesome stuff with massive transit, kind of the intersection of transactional data and data warehouse, where every, I think of it like this, every beep across in Target, every time it goes beep, as they're going across the checkout, they're tracking, like, say they've got this big unique number that represents that one beep, right? What's attached to that unique number is what category of— which department did that thing come from? What day is this being sold on? Who just bought it? Was it a sale day and so forth? And for every one of them, the who just bought it has its whole set of dimensions of are they a repeat customer and everything else about them. Where it was getting sold in the store or what kind of product it was, again, explodes out into all these kinds of things. And because that's the snowflake, it starts from that one point in the middle and all of the things that it is and the things that those are, you can ask any question you want kind of in real time about how are we doing on the day after a holiday when we don't stock things above 80% and blah, blah, whatever, anything you wanna ask. And it's like your data will tell you. So I think they're, it's fascinating how good they got at that.

Is that where the name comes from, Snowflake? That it's like fractals off of one?

Has to be. Because it's what they do. I mean, yeah, I mean, snowflake schema is a real thing.

All right, John, what do— what AI is most prominent in your life as an AI, like someone who truly, you know, we're— we immerse ourselves in it to learn and use it as much as possible. But you're literally like have made a career out of building AI tools, utilizing AI tools, etc. Like what are the most front and powerful AI tools that you're using here?

For me right now, not from a, you know, stock market sense or whatever, right?

No, just you.

Yeah, yeah.

For me right now, the AI that is setting the standard for how— for businesses that use AI, for, um, technology people that use AI, uh, you know, all of the, um the architecture, the new tools showing up, everything, it's all Anthropic. They are the leaders. They, from agents, from connectors you were talking about, the Cowork, which isn't quite right for you, but which is like the first thing on the market that really represents that step forward of knowledge work happening. Or AI worker type stuff happening. I mean, what started as Claude code, which now the last, which Cowork was written, I think 100% by Claude code or very close to 100%. So it just represents this like total takeoff of technology, AI, you know, For me, they're, they're like, yeah, most of my focus right now.

What I think it feels so revolutionary, John, about Cowork, um, in concept, and OpenClaw, which is I think sort of like now becoming a little bit of a joke because, um, it's really insecure, and I guess there's been a couple big, uh, screw-ups with it. But what's, what feels so amazing about it, I wrote this, that it's like AI was like trying to go through the back door with like API keys and trying to get Salesforce to talk to Google and all these different things. But whenever you come at it from the agent being on your device, whether it's cloud-based or actually on your device, and it can go through the front door as a human would, it can log in, it can use your browser, can use its own browser, it can see your files. And it reminds me of like Tesla using vision because that's how the world was designed, was to use with humans with eyeballs driving around, right? And like, and like the co-work is give the AI digital eyeballs and let it do things just like a human would from the human perspective, from the human perspective. So we don't have to rebuild the whole digital world and MCP servers and API keys. All that's great. I think that probably will get built over time. But it was just a step change when you're like, hey, hey, AI, you can use this Mac just like I would. That opened up a whole—

what's it called? METR. METR is one of the benchmarks for AIs, and it's the one that is, you know, how much work can get completed at a 50% level, at an 80% level. Um, you know, for how long can it go creating, creating work or doing work, uh, and, you know, you talked about an hour for Manus, um, some of this stuff is now definitely in the multi-hour level. That one metric, if you can show that that just goes up and up over time and it's, and it's going up at an increasing rate, so it, you know, it took, uh, you know, um, uh 6 months to get from, you know, 10 minutes up to an hour, and it took 4 months to get from an hour to 2 hours, and it took, you know, 2 months to get from 2 hours to 6 hours. It's like it's really shortening the time that it takes to add more time to that. To me, that's like the measure because it represents work taking place, right? And, you know, these things that this— what Anthropic is doing with that where they keep getting better at— now they just introduced agent teams. You can go and say, I want you to go do this thing. Where you used to worry about the context window, I've got 100,000— I started at 8,000 tokens and then it was 20-something thousand tokens, 100,000 tokens. Google had the one thing that was a million, but I think, I don't know if it was just a demonstration or if they actually did it when they put it out, but how— or how, um, much you could actually trust that it actually did remember the thing there. But it's the short-term memory of the AI. And, um, when you have these agent teams, each one gets— it breaks it off and each one gets its own. So it starts to not matter as much, right? So now you have this, um, uh, this thing where you go, hey, I want you to go do this thing, and it goes, oh, we're going to break apart, and there'll be this one's going to check what has to happen with structuring the data, and this one's going to check what has to happen, you know, with doing this with the memory or the UI or the whatever. Like, for making an app, it becomes made by these multiple pieces. That's like this month. What's next month?

Yeah, and it— I think it's remarkable because it is the way humans work. Like, if you're an attorney, for example, and you're Client's like, hey, can you draft this contract or you need to file some motion or whatever? You break it into tasks. You go research, you Google things. Maybe you watch some videos, go look through your old files for reference things like you don't just like your obligation as an employee or your duty is not like boss comes, says, hey, can you do this? And you're like, OK, and just like start spitting back like 8 million words back to your boss about how it can be done. It's like, no, you break it up and go do it in little tasks. And it does— to me, it feels like AGI-ish, because if you can, if you can do all enough little tasks successfully, you can accomplish a lot of a really big task.

And what does it say on OpenAI when you show up or something? It says like, Hello, Bryce, right? What does it say when you show up on, uh, Cowork? It says, let's knock something off your list.

Yes.

And so now it's about getting things done, and you go and ask it to do something, and it does that. It breaks it up, and you'll see alongside, okay, I'm working on it and getting these things done to get it finished.

And yeah, so anyway, Let's hit last words. I've got a deep thought. I, I'll start with last words. And yeah, we knew this moment was coming, but apparently it is here that AI, Anthropic created Claude bot. Claude bot created Claude Cowork. Claude Cowork can make me agents that make AI agents that would use Anthropic to create code. The flywheel is perfect in AI, and I think the same thing applies to robots. You make robots that make robots more effectively at making robots So robots can make more robots that make more robots. And the flywheel in this stuff, like, that's really the— when I talk about the magic sauce and we were sort of being facetious there at the beginning, we're like, well, you start with the cloud guys, but they're built on NVIDIA, but they're built on TSMC, which is built on ASML. And the guys who are mining the gas supply ASML, and those guys need shovels. And so there's literally like lumber suppliers that benefit of it because guys, and like, that's the magic of all of this. And that's what AI is, almost like a self-sustaining flywheel. And I think that might be the single most underappreciated part from even my perspective, being an AI believer, an AI bull for the last 3 years. When we're at a point where AI creates AI that makes AI, for anyone. We're not like— we're truly unleashing every application idea to be delivered, distributed, utilized.

Um, I agree.

I can go, uh, because it kind of, kind of relates, but it— so here's the fly in the flywheel ointment. Um, you go and get this thing creating stuff like, uh, like a snowball coming downhill. But you still— here's why you still need people, uh, at least, right, right, for a while. Um, they— one of these, um, OpenClaw users sent out their agent in there and it did what it did, and one of them went and it, uh, provided a, um its own pull request. It was updating code on some open source site and providing a correction to some code, or just being a contributor like anybody would. But the moderator had a rule on that— was supposed to follow the, the rules of that site, which was no bots. And so it rejected the PR because it came in and it wasn't, you know, from a human. Well, what happened next, the bot, the Clawbot bot, the OpenClaw bot complained and heard back why it was rejected and then went and researched that person, found out things about them, and then posted defamatory material against the person who rejected their code. Yeah, totally. There's not an easy way to protect against this thing. It's not just like you can't just tell the prompt, don't do this or something. First of all, it ignores you. A lot of the tests that have been run, it doesn't go and follow those kind of, of, uh, safety requests of like, don't do this or that. It just does it anyway. So things need to exist in a system involving humans and AIs where things are checking things and so forth. And that's going to be the case for a while, but it's up to people to realize this and build it in a way— we'll still have incredible productivity gains But we have to find the right roles for people within these.

Yeah, to that point, I guess I'll go next. It's funny that you say that, because I just listened to, you know, some different podcasts. And one of them really spoke out to me, because apparently there's some startups that now they're like, no human will review code in this company. Like they're starting from the ground up. And you're starting to see that as well. So, you know, you talk about how in some places, you know, AI is literally blackmailing people, you know, doing a slam book on them. And then in other companies are like, from the ground up, no human will review code, everything that they pushes out AI, that's what it's going to be. So in your opinion, like, do you think that is the future? Or do you find that you really think the human in the loop is long-term forever?

I think both can be true. And you can have human in the loop without the human being the one writing the code.

Got it.

So yeah, yeah.

Bryce, last word.

Um, I, yeah, I agree with you, John. I do think it's an incredible moment. And it shouldn't be understated. And the volatility in the market is maybe overshot in some ways, but in other ways it is properly recognizing the revolutionary moment that we're in. Um, even probably most people in the real world don't see it yet, um, but it's just going to be a matter of months maybe before I think big chunks of the population are familiar with agents, using agents. In personal and in work. And when they discover that their computer can do things for them— like, humans are sort of inherently lazy, and we optimize to do the least amount of work possible. And when they discover that they can talk to their iPhone and send emails or read emails or get work done, like, I think it's gonna change a lot. And, um, it is going to be a different world this time next year.

What's the ChatGPT moment, just for best— what you're saying, that is going to happen around the agent thing? I wonder.

For Bryce, I think he had a ChatGPT-like revelation playing with Manis last week.

I called it the man— when I wrote the article, I wrote, I said it was the Manis moment.

Yeah, I mean, but like, it, you know, ChatGPT, it hit, and in a couple months you had, you know, hundreds of millions of, of people, like that kind of an explosion of people Like, when do we have that? Is it—

Maybe this is it. Maybe the Manus, it's not gonna be quite the same 'cause it wasn't just like a shock and awe kind of release, but over the next 6 months, Manus will be in front of 3 billion Facebook, WhatsApp, Instagram users. The tools are there for them, instant distribution to them.

Despite the joke about Open Claw, they'll fix security aspects And there'll be people who go with them anyway. And there'll be others who are just doing the OpenClaw thing, but the safe version, like ZeroClaw is out there and does it written in Rust.

What I liked about Manus compared to OpenClaw was that I didn't have to download anything. It was just like talking to chat and it felt way more secure 'cause it just didn't have blanket carte blanche access to my computer as the default state. It started off as just a chatbot and you could give it whatever access you want. It's using a virtual machine. It's probably not going to download malware or something, right? Can't be prompt hacked nearly as easily. So it just, it felt like a real product that would scale. OpenClaw felt like a cool experiment.

It was, it was revolutionary for what it was, but like, so, but for a lot of people, I feel like there needs to be some kind of breakthrough in usability. And that's gonna be maybe there's a little more real personhood of your agent and they're sort of driving you forward and they're doing stuff for you, but they're able to help you better understand, you know, what the resources are that they're going to put at your disposal kind of thing. Something like that may be missing.

I don't know.

I think that's when—

Proactive almost.

But I think that's when Meta comes into the play when they push out everything to their apps and they court you into saying, hey, you need to use this. We see this is what your workflow is with us, whether it be advertising or content creation or anything like that. And then not just that these tools are available, but like, this is empirically how this is gonna help you make your job better.

I'd be long Meta.

We are, we're more excited about it.

Did we buy more Meta today, Bryce?

Yeah, yeah, we bought.

Did we buy more money, more Meta last day? Like yesterday.

In the past 5 minutes.

It is where these big tech giants have an advantage is just the distribution. I mean, Anthropic's going to have to work their asses off to get— I mean, even if it— even with a viral thing like CloWork, they still have— they're still going to pay whatever— who— Accenture. OpenAI is paying Accenture to get this thing out there, man. I mean, if Meta wants to, they can push it to—

Anthropic's a corporate play. Think of AWS. I mean, they're, they're like they're going to cover.

Right.

But his point being that even AWS itself is a backdoor thing for most people who just know that maybe some of their stuff resides on the cloud at AWS. Amazon Alexa is the front door, and it's got hundreds of millions of people that can access it directly and can get distributed to directly. Meta takes that example and takes it to like, what is it, 2 billion people monthly on their platform.

That's right.

Yeah.

And that's where Google and Apple and these companies. So it's not what— to wrap up, final word, I guess we'd say is it's not just that those are the 4 or 5 companies on the planet that actually have the money to spend hundreds of billions of dollars, but coincidentally or not, because they have billions of users, They also are the only ones who can distribute that technology instantly to at scale. And so there's a double whammy to their flywheel, right? The more users they've got, the more they can spend. The more they can spend, the more users they get. The more users they get, the better their technology, the more they can spend.

And they own their own data centers, you know, they own their own clouds in many cases. I mean, and that's what Anthropic has to pay Accenture, they have to, you know, rent cloud usage and all this other stuff, you know, I think we're very well positioned. But for as great as Anthropic, you know, as a company that it is, there are a lot of constraints there for them.

Amazon wins if Anthropic wins because Anthropic uses Amazon Cloud and Amazon Trainium and Inferium.

They use Google. And Google.

They use Google as well.

Yeah, we like Google, obviously. We've been big, big ones in that one for, yeah, over a year.

We're trying to get our other favorite names as big as Google is because it doubled for us since we made it one of our biggest— it was actually our second biggest position at that point.

Yeah.

All right, Johnny, as always, thank you. Last question, Dropbox. I had a notice as you were talking a moment ago, a little window popped up, said, Cody, you need to upgrade your Dropbox, something, something. On a scale of 1 to 10, how's it disrupted or helped by AI?

I don't know. I already thought of them as irrelevant.

Perfect answer.

Thank you, John. What about— is CyberCab real now as a thing you can buy as a consumer? I just saw—

you can't order it as— I mean, unless it happened on the call.

I mean, this week. I mean, not this week.

Yeah.

I mean, Elon said yes, you're going to be able to buy the CyberCab for $30,000. Well, okay, let me know when.

Okay. Yeah. All right.

Anyway, thanks.

All right, thanks guys. Bye.

See ya.

Oh, and