A (1:24)

But it needs to be well targeted.

B (1:25)

This episode is brought to you by openclaw.

A (1:27)

Yes, likely.

D (1:29)

Does openclaw have any money to pay us? Dave, what's going like, I don't actually understand what the openclaw foundation does.

C (1:34)

Right. With the hard hitting news. I love it. Sam, Good. Good instincts here.

A (1:38)

Yeah. Wait. Way to moderate, Sam.

B (1:40)

The openclaw Foundation. Do you want me to answer?

D (1:43)

Is it a nonprofit? Is it a Cayman entity?

B (1:46)

Yes, it's a nonprofit. It is a Delaware nonprofit.

D (1:49)

Delaware nonprofit. Fancy.

B (1:50)

It's a Delaware nonprofit. We are chartered as of last. It's actually news I haven't published, so I should probably put it out there.

D (1:58)

Will there be at least one fishing representative on the board? Fishing because it's claws.

B (2:04)

Oh, no. Yes. Like a main lobsterman fisherman.

D (2:07)

Kyle could do it. He could represent the fish. I speak for the fish about this.

A (2:13)

Okay.

C (2:13)

Before we. Before we devolve, before having begun, this will be a claw episode because we're going to talk about Nvidia's GTC where many of us were in attendance, which sort of like the AI super bowl and the claws were a big theme. So we're going to talk about that. We're talking about some awesome news. Out of the awesome news, out of the important news, out of the information about Apple and Vibe coding apps, which I think is under appreciated and interesting.

B (2:41)

Yes, it is.

C (2:42)

Those topics will take up a lot of time. I am sure we will get to more. But first, in a little job ad, I suppose. Sam, you posted you're looking for a Master of Bots. Can you talk about this?

D (2:55)

A Master of Bots?

B (2:56)

Mom, I thought you were a Master of bots.

D (2:58)

I am, but I'm too busy, like, so I. This is like, I think.

B (3:02)

What else are you doing with your life?

D (3:03)

You'd be surprised. Playing tennis, lifting weights.

C (3:06)

Sam joined a tennis league.

A (3:08)

Oh my God.

B (3:09)

Oh, God.

D (3:10)

Here's the upshot is you remember how like back in the day you used to have like a sysadmin at your company? Remember those days? I think there's a new role which is the rise of the re. Rise of the system, which is going to be called the Master of Bots.

C (3:23)

But if it's all dark factories, what is there for the master to do?

D (3:27)

The master's job is just like literally to manage the bots, deploy new bots, sit down with people who don't the fuck they're doing and help them build the bots, maybe build some software. It's just. It's just new.

A (3:39)

I talked about this like 10 episodes ago and we called that the.

C (3:43)

Yeah, I think you did.

A (3:44)

Thank you. The Chief Agent Officer or something like that.

B (3:48)

I like Master of Claws, Master Claw

A (3:51)

Master or Chief of Agents. We call that Chief of Agents Claw Master is good.

C (3:55)

A coow.

D (3:56)

You could also call it. You could call it AI hr. Like, I want an HR manager. I want like a mo. An H. It's not hr, It's. It's a R. Air.

A (4:07)

Okay, Sam's not the branding person on this pod, so let's not do that.

D (4:11)

It's just like, basically like for every 100 software engineering jobs, there's going to be one master of bots job created. Right? It'll be, you know, 100x downshift.

C (4:19)

Okay, that reminds me, we have to talk about layoffs to third topic.

D (4:22)

Yeah, but it's like I kind of want a person who has a CS degree because it's just proves, you know, how computers work to some degree and like is really into the bots and can think about security and like just basically packages things for people to spit. Because it's like, it's just an. I can do it. I Just have other shit to do.

C (4:39)

Do you guys have bot anxiety?

D (4:41)

So I know bots are great.

C (4:43)

No, no, I. When I can't access my bot, I get anxious. So I had a long drive to San Jose in traffic and I wanted to do work on my Claude code. I wanted to update this report it's generating for me and someone called me and I was like, oh, I wonder how their last story, what the story around that was. And I could not because I can't use it on mobile.

B (5:09)

I thought you could.

C (5:10)

Now I can. I can use my Claude code thingy on mobile.

D (5:14)

Yeah, there's multiple ways to do it. Oh no, there's multiple ways to do it. You can either just boot a remote session, which is easy, or you can SSH in like through Claude's interface.

C (5:25)

Okay, well maybe we can figure that out.

B (5:26)

Jess, we can get your claw doing this for you. It's no problem.

C (5:29)

You know, I have this. I.

D (5:31)

But Jess is rich, so she uses Claude. Rich people use Claude.

B (5:36)

She uses Lux. She used luxury AI.

D (5:38)

Poor people use the. Like the, the. The claw. The claw. It's like kind of.

C (5:42)

I actually had a meeting today out of how to de luxury my AI and how to expand it to my team at a more cost effective. So I'm currently narrowly in the luxury AI window and then I'm going to move into the widely deployed cheap AI window.

D (5:55)

Well, it's funny, it's like that's actually the real upgrade. It's not how many tokens you get, it's the token. The thing is like when you're like a low level employee, you get claws and at a certain point you're granted luxury Claude.

A (6:07)

When you get promoted.

D (6:08)

You've promoted to have clawed now.

C (6:11)

No, but then you do too much and you have to go back. But anyway, so I, but I did not like this about my. I did not like that anxiety I had. It seems like I just needed to take a deep breath and remember.

A (6:20)

Well, it's not only anxiety, Jess, but like there's actual now data coming out that people have been measuring people's. Not only their memory, but their problem solving, like their constructive thinking and people. The more people are using all these AI tools, of course, the more that those things are going, I, I feel like we're going to need some sort of brain gem that keeps us sharp because we're outsourcing all of these hard thinking things to all of our agents.

D (6:44)

They make you play Sudoku while you wait for your shit to get done.

C (6:47)

I was going to say Sudoku or crossword puzzles.

D (6:51)

But I love Jess. Jess caught me. I like built an entire app flying home, like last week. It was great.

C (6:56)

I was just looking at the back of his screen. I'm like, that thing is just running at like 10,000ft or wherever we were.

B (7:02)

Yeah.

C (7:03)

And I was like, that's pretty impressive. Like that. That was like kind of wild, to be honest.

B (7:07)

I've actually thought about that. Vibe coding on planes might actually be a thing. Just like, buy. Buy it. Buy a plane ticket and coach and just like vibe code. Just go places just to vibe.

D (7:19)

Dave, this is an old idea that I will credit. Ricky Van Veen and I were like, co developing on a walk once, which is. Here's the idea is co working spaces that are airplane fuselages that don't go anywhere, right? So you buy a ticket and like, it's a super dense thing. Like, think about it. You can fit so many people in a small space. You have a stewardess to sound. You just sit here. On your calendar, it says you're in

B (7:43)

the air, but are you like forced to stay in the chair for like the time period you've rented it and you can't get up?

D (7:48)

No. You go bathroom. You can go to the bathroom only. But it's. Think about it. If you're on an airplane, it's the only time that you're allowed even still to set. Like, I'm not available because I'm flying though. You are available, right. And then you can just be left alone. It can be like 2pm and you can watch a movie. You can vibe code. You have someone bring you food and it's incredibly space efficient and you can have super fast WI fi. So like, to me, the whole. The innovation is airplanes that don't go anywhere, right. Is going to be like the co working space of the future. You could even block it on your calendar. You could. You could buy it as like, I'm flying to Tokyo and that's like a certain number of hours.

A (8:23)

Or I'm flying to like, buy it hourly.

B (8:26)

Yeah.

D (8:26)

Like, oh, I bought a Tokyo ticket.

B (8:28)

You know, part of the productivity is the pressure change.

D (8:32)

You can make that happen.

B (8:33)

You could like, you like, pressurize the fuselage.

D (8:36)

No, it's the opposite. You're pressurized to a very high altitude. So you make you slightly stupider and remove some air and then you get better work done.

B (8:43)

But.

D (8:43)

And you think everything's a good idea because you're loopy, Right?

B (8:46)

Pretty good.

C (8:47)

This is not the best idea. Not the best idea.

B (8:50)

Or you could just go get On a plane and just fly somewhere just to code.

A (8:54)

Let's go back to the clause. Jess, we're going back.

C (8:56)

Thank you, Brit. Okay, so keep us on track. I would say a couple years ago this became a big thing. I mean, it was always a thing, but it became like a big thing.

D (9:05)

I thought GDC was Game Developer Conference.

B (9:08)

It, it is.

A (9:09)

Gtc, gdc, all the acronyms.

B (9:13)

You guys, what is the T. Wait, GTC, right?

A (9:16)

GTC versus GDC.

D (9:18)

Here's the thing. It's all, it's all GPUs. It doesn't really matter whether it's GDC or GTC.

C (9:24)

It's true, it's true. Although there are a lot of people trying to sell TPUs at Nvidia's GTC, let me tell you, which is obviously Google's competitor. But this was the first time I ever went. I, I just felt left out. The information we set, we're sending like three reporters and three business people and I'm like, you know, I just kind of want to go.

D (9:41)

I thought you looked good. I like the information read. You were supporting the brand.

C (9:44)

I had such a great time. Guys, I want to hear about Dave's. I was not at the keynote. I just went and I like had a couple meetings and then I've walked the floor, which I haven't done in years since like ces and I just went up to random people and I think I like reporting is what I'm learning. But I thought it was fascinating. There were so many people, so many companies.

B (10:08)

It's really unbelievable how big it is.

C (10:10)

So much excitement and I will just say many fans of more or less. I took seven selfies and I don't think it was because of the information's hard hitting business coverage. I just don't. I think it's this guy's.

D (10:23)

So this whole narrative that like you're. You'll get positive feedback from more or less is wrong.

C (10:28)

I know I have to zip it. I mean it was very valid. Posted them online. I was like mortified. But anyway, so I thought it was great. I was mostly there for a Jensen press conference and I will give my recollections on that. And then I want to hear what Dave thought the keynote was like and talk about the claw. All the claw news. But you know, I haven't seen, I've seen Jensen speak and I like shaken his hand before and stuff. But like this man loves a press conference. Oh yeah, he was relishing it. Every second he was working the room. He was like kind of making fun of people at some point, but like it. I know Sam talks about like narrative capitalism and these CEOs who are like winning on narrative and storytelling and like the cult of personality. And obviously Nvidia is also winning because of the gpu. I'm not. That is why they're winning. But I was a little surprised of just how he was also kind of cut from the same cloth. Like, he is really embracing this role as AI's Storyteller in Chief and just relishes it. I mean, he actually said during the press conference, oh, I want to keep talking to you guys, I don't want to go to my next meeting. I mean, he kept. So that was it. But I don't know. You guys have never gone to these as reporters. This was also true at Apple events and still is. The audience is packed with a wide variety of people. But the Fawning over these CEOs is also a little bit much. And there are people who presented him with awards. There's a good 20 minutes of selfie taking afterwards. And I mean, obviously he is a rock star in like many dimensions, but he was obsessed with the fact that he had been crowned the inference king, which of course is a big deal because people used to make the claim that Nvidia was weaker at inference. Then they bought Grok and obviously blah, blah, blah. But like he talked about that award maybe a dozen times during this press conference too. So anyway, he's a real. Not only is Nvidia an important technology company, but like he is relishing this moment.

D (12:40)

Why not give it to him?

C (12:41)

Soaking it all in. It's not good or bad, it's just there are a lot of. You cannot get Sundar on a stage. You can't even really get Mark on a stage or Demis on a stage or, you know, like, I mean, it's interesting.

B (12:53)

Well, I think that it's an interesting, like power to the guy, right? Like he kind of toiled away building this technology, like fundamental chip technology company for a lot of years before this current thing happened. Right? So in a way it's a new thing, right? And it's kind of the payoff of like a really long journey.

C (13:17)

Totally, totally. And I will say, like the other things I learned just from like conversations there is like, you know, the ecosystem is strong. Like the narrative, right, is they have the dominant piece of hardware for this AI revolution. But like, I got this sense that they also have been working with so many companies and so many industries in a really hands on way to like, help them and build alongside them.

B (13:44)

It's a really impressive company on that front.

C (13:46)

Sometimes you hear complaints because, like, no one likes to be beholden to one supplier for anything. But, like, the flip side of that is you really get a sense of, like, how they earned it and, like, are deeply, deeply integrated with, like, all of these partners, which I think is important to understand for the question of, like, you know, how easy will it be for people to switch and so forth.

B (14:09)

So, yeah, I mean, I sat there, Jess, and I was like, we in the tech business spend so much time talking about the AI application layer at the end of the day, right? Like, or whatever you want to call it, the foundation models and then the applications on top of them. But there's this other company called Nvidia, which is the actual company that is the platform that we're all building on top of. And we're so abstracted away from it now that, yes, we talk about it, yes, you talk about the public stock and how it's been ripping and all these kinds of things, but the Nvidia conversation is sort of so abstracted from the everyday tech talk that it is really shocking to sit there. I, I, I was just kind of in awe of the whole thing.

C (14:59)

Like, yeah, talk about the keynote, Dave, and what stood out.

B (15:02)

Yeah, it was very long. I'm very, I'm very impressed by him.

A (15:05)

He was three hours, first of all. Definitely. Speaking of a man who likes to speak. That's a lot of energy to put out for three hours.

C (15:14)

Apparently he needs a lot of snacks. I also learned that when I was.

A (15:17)

If anyone wants to watch it, I was. Brit Morin is featured at 2 hours and 21 minutes. So just fast forward to that.

D (15:23)

What are you featured for?

A (15:25)

Clock on.

B (15:25)

I mean, it's really impressive that the guy can go three hours with no notes. Like, got to give him props.

D (15:31)

You couldn't speak for three hours with no notes. I could.

A (15:34)

Dave could definitely speak for three hours.

C (15:37)

I think this one point is a little overrated. Like, he also was speaking extemporaneous. Like, you could tell he was like, it wasn't super polished every word, which was fine.

B (15:47)

But yeah, yeah, sure. I mean, I think he put puts a lot of work into it, but I guess I was just struck by what I said that, you know, here's this company that happened to build the right technology for this moment. It's been a long time coming, but the sheer level of, I mean, $1 trillion of demand for their chips over the next year, like, this is a really Crazy moment. And I think that was also this conference this year. Just the energy of the whole thing was really quite impressive, but also says something about where we're at in this moment as an industry and as a world. And I think they're definitely at the center of it. And at the same time I was like, wow, we don't talk about this enough. And yet there's still another company underneath this company. Right. There's TSMC and the power.

D (16:43)

Don't forget the power.

B (16:45)

Yeah. And how much we don't talk about these things in the context of the AI revolution. So I was just really struck by how big it is. I mean, the entire Shark Stadium was full, like whatever. I don't know what the capacity is, but something on the order of 25,000 people completely sold out every single seat. Somebody compared it to Taylor Swift for men. Actually that was pretty funny.

A (17:12)

For straight men, Dave. Straight men, Yeah.

B (17:14)

I don't know. There's a lot of gay men there too.

A (17:16)

I'm saying the gay men are definitely swifties.

C (17:18)

Well, let's talk a bit about. So one of the big announcements and Brett, want to bring you in on this too. It was Nemo Claw. So obviously we know Dave is on the board of the Open Claw foundation and has been really involved in the development of that.

B (17:33)

And my understanding I'm now like calling myself. I'm like the consigliere of openclaw. It seems like.

C (17:39)

I don't know if you want to call yourself that. We'll have to think about whether that's a good self anointed title.

A (17:44)

It's a title.

B (17:45)

Chief Claw do a lot of things.

C (17:48)

Yeah, but so Jensen was very excited and Peter, the Open Cloth founder was there, right?

B (17:54)

Yeah, yeah, Peter and I were there and one of our chief maintainers. So there were three of us representing openclaw.

C (18:02)

So talk about Nemo Claw.

B (18:04)

Yeah. So, you know, Nvidia has been really helpful to OpenClaw over the last many weeks. They've actually brought engineering resources to the bear. You know, one of the hardest problems that we've had is getting really talented engineering talent involved in. In the world of maintaining openclaw. As it's scaled up so fast, it's been just like really chaotic. I mean, I've been comparing this. The only other time I've been kind of in the center of the storm like this was when I was working on Facebook Platform at Facebook in 2006, 2007. And the speed at which this has grown, I think kind of outstrips anybody's ability to execute. And the hardest part has been maintaining all of the, all the pull requests that are coming in as well as the security reports. And so Jensen and the team at Nvidia contributed engineering resources. One of the big things with openclaw, there's been a, you know, a narrative out there that it's not secure, that it's not safe. And so we wanted to work on that with Nvidia. And so over the last couple weeks we put together these two things. One's called Nemo Claw, the other one is called Open Shell. Nemo Claw is a reference implementation that makes it easy with one command just to set up OpenClaw on prem with really good security guardrails, privacy controls and importantly support for local models. A lot of folks that are setting this up in the enterprise want to be able to run open source models that could be like Nemotron, Nvidia's local model, or it could be Minimax or whatever you want, whatever local models you want to run so that your agent can run entirely on prem and no data leaves the building. And so this is just a really good reference implementation that enables that. And then you can go and customize it however you want, but it'll work on the whole Nvidia stack, makes it easy for anyone running Nvidia infrastructure. And then the second thing is this new thing we created called Open Shell, which is basically the secure sandbox. It's like a runtime that runs underneath Nemo Claw or any agent. It just enables you to sandbox it into its own environment. Jensen was happy enough with it that he made it a meaningful part of his keynote. And so that was kind of the delight for us was we weren't sure how big of a part of the keynote this might be. We were actually two and a half hours in going, oh, is it just going to be one slide? And ended up being, you know, 20, 30 minutes of the keynote. And they made it this really wonderful storytelling about the openclaw community. And Jensen said that it was openclaw's maybe the most important piece of software created since Linux or Windows and many other very nice things, you know, is great, great for openclaw, great for the movement and frankly, you know, I don't think would have been possible if we hadn't have maintained this as an open source project. So it's exciting.

C (21:04)

I heard him say he thinks it's like I think he said this or someone told me that he believes this. It's sort of like the chat GPT moment for agents on the consumer side.

B (21:15)

Yeah. You know, his opening slide I thought was instructive about why Jensen and Nvidia think this is an important moment. The opening slide showed the chart that everybody's seen now of the Open Claw, the number of stars on GitHub going fully vertical all the way up to this guy.

D (21:34)

Just. I just got this in my inbox 0 minutes ago from code 2. Can you see it? Yeah.

B (21:39)

So that's the chart and Jensen framed it as the inflection or sorry, the inference inflection point. And I think that's true. Right. Like what you're seeing is the application of inference to a lot of different domains equals extremely vertical lines. And in this case it's the operating system for agents. And that OpenClaw is or has rapidly become the operating system for the agentic economy. And we're here now and it's happening very quickly and everyone's adopting it. You know, one of a big stream of my meetings over the last few days, GTC were all of the different local sort of open source models, your minimaxes. We've got a lot of people from China, Baidu, Xiaomei, many, many different companies that are all developing, some are developing custom models just for openclaw, some are integrating into their clouds, some are doing other interesting things. But you know, we've got global interest from, you know, companies all over the world, not just American, Chinese, Europe, all over the place. So it's been really interesting.

D (22:59)

Can you back up and describe what's the software? And this is a question I ask is like what, what is really Open Clock is to me, again we think at a high level it's persistence and identity on some level. Right. But it doesn't, unlike a Linux or everything, it's not like a heavy piece of software. So when you think about like the few, how you define even the guardrails of like what Open Claw is, what is it? I mean, you know, it kind of blew up because it had some persistent memory and like did it. I mean it's very useful and then like bridge to Telegram and so it made it really easy for less technical people to like consistently interact and build skills. There's not a lot of like complexity to that and that's not necessarily a criticism. I'm just trying to understand like when you really got to the heart of it, what is this thing other than like kind of a shim?

B (23:46)

Yeah, I mean I think that one way to describe it is that it's a personal AI agent that runs on your computer. You can get More technical about it and say it's a harness. You know, a big part of the, what everybody's talking about right now is building harnesses or orchestrators that orchestrate all of these actions, tool calls, skills that AIs can employ. And so it's a really well done one of those, but just to push you.

D (24:16)

And again, this isn't a criticism. I'm just trying to understand how you think about it is like this is

B (24:20)

Sam's just to push you. I'm not criticizing, but I am, I'm not criticizing.

D (24:23)

I'm actually like, I just like I'm trying to think about how you intellectualize it.

A (24:26)

Learning, Dave. Sam's learning.

D (24:27)

Like I can obviously like publish on GitHub repo like a bunch of text files, right? And suck it into openclaw or Claude or whatever. It's like very easy. Like there's no, there's very little to it.

B (24:38)

Openclaw will run.

D (24:39)

I understand why Nvidia loves it because as long as it's inferenced somehow, they're getting paid, right? So like six levels down, who cares what model you're running or how you're arbing it or whatever.

C (24:50)

But like, by the way, you'd rather not just have four major customers. If you're Nvidia, you'd rather have like 4 million?

D (24:56)

Sure, you can have as many. Like it doesn't matter. Like you can have open router, you can route between them. Like it's just like, it's just all power and matrix. Multiple application on some level, right? And so like, I'm just kind of curious like, you know, you think about like what an operating system was, right? And is like versus what this is like, you know, I'll give you an example. It's like someone was pointing out that there used to be this whole narrative about taking PC games and porting them to iOS or to OS X and how impossible it was and whatever. And now like you can just do it on the fly. This is like that only like it truly doesn't matter. It's just like a few text files you scrape in and it all works, right? Like, and so I guess the question for me, and like, by the way, it's just pushing tokens at effectively like anyone, right? And so I guess the question for me is like, what when you think about the future of this, is it identity management, Is it enterprise? In case, like, what is like, what is it other than kind of like a ship?

B (25:49)

I mean, you could call an operating system? I mean, what was an operating system? Right. Like what was Windows, what was macOS that we are using right now? Right. Like, it was another shim. It was like an abstraction layer that enabled you to do things above the assembly layer to compile things that the CPU could understand. It was a UI layer that enabled you to understand what files and folders were in a way that was visual. It was a bunch of other, like, runtimes and things that enabled different use cases. So, like, you could call any of these things shims, right? It's like how you put them together. And like, I think that what was so unique about what happened here with openclaw is that he made really interesting technical choices, right? It's like this simple node process that's running on your machine. It's largely using TypeScript, which is like one of the best languages with the most support, right? Like, the node community is enormous. It connect to any LLM via API. I think one of the smarter things that he did was this sort of skill slash plugin system, and that as you start using it, it can improve itself. Like, I think this is Sriram actually, who works on AI in the White House, was talking to me about this yesterday. One of the things he thinks is the most unique is that it's one of the first one of these systems that you can actually teach it to improve itself. And that that's a very unique property of it. And it has memory. Like, I think one of the reasons I was drawn to it straight out of the gate was that it reads and writes files just like UNIX as memory, and that those files stay on my computer. I can use them as raw material for essays or podcasts or whatever it is that I'm working on. Like, I now have my files on my computer and I can mix and remix them very easily. And it's constantly writing those and improving them and changing them and adding new ones. Like, that's like a new thing that nobody had done before. It can spawn sub agents to do parallel work. And then like you said, the interface, I think, was actually a breakthrough to Jess's point right at the beginning, which is like she wanted to do something on, you know, when she was away from her computer. And so giving people the ability to use it over text message, whether that was telegram or, you know, imessage. And I actually think one of the very simple but nuanced things that was different is that the way that compaction and memory is written in this version of this is really simple. It automatically, you know, it runs a context window. Once the context window fills up to a certain level, it saves everything to a memory file.

D (28:41)

Sure.

B (28:41)

Which you can write that a bunch of different ways.

D (28:44)

Claude does all this as well. The only thing it doesn't do is let you switch which backend you're using.

B (28:48)

Yeah. And like, but I think that's important.

D (28:51)

No, I get it. I just like when you look at something like Nanoclaw, like, which is a separate project, like, I just wonder how it evolves. Right. Because like, to me it's like, it's kind of like imagine a future. Imagine you had oss, but it was unbelievably, once shown what to do, easy to make your own os. Like, how does that ecosystem work is like almost the question.

B (29:12)

I think it's just like Ubuntu and there's like many flavors of Linux. Right. And for some, for one reason or the other, these like massive communities generate around, you know, different types of Linux. Like there's lots of flavors of Linux, but when you boot up a new, a new droplet on Digital Ocean, you have to choose between Ubuntu and all these other flavors.

C (29:32)

Okay, you've lost me at droplets. You've lost me at droplets.

A (29:36)

Guys, I'm sorry, this just chooses Ubuntu.

D (29:39)

You actually are literally using one right now. You don't even realize.

C (29:41)

I know I'm using so many droplets and tokens and all of these things, but. Breaking news from the information. Sorry I interrupt this program.

A (29:51)

Right turn.

D (29:52)

I'll call you later. Dave.

C (29:53)

Guys, inside Meta. A Rogue AI agent triggers security alert. We just published another story. These have happened before.

A (30:02)

Was it a claw or a different agent?

C (30:04)

It was an in house tool similar to openclaw. To be clear, nothing was exposed to publicly. But I'm going to read this because it's very technical. A rogue AI agent triggered a major security alert at Meta by taking action without approval. That led to exposure of sensitive information and user data to Meta employees who

D (30:27)

could sound like an intern. You know, many interns have done that.

C (30:30)

Is this happening at every company in America right now?

B (30:33)

I know that's.

A (30:34)

They weren't using Nemo Claw because that was. That's the secure enterprise.

C (30:38)

Let Dave comment. And then, Brit, I want your take on all this.

B (30:40)

I said Meta is, is releasing a version of this story every couple of weeks and so it's hard not to

D (30:46)

release it just broke it.

C (30:48)

They are not. They are fundamentally not releasing the fact that they gave employees access to data they shouldn't have inadvertently. They're not happy about this. But you think they're Trying to make open claw seem less secure because whatever they're going to do is going to be more secure.

B (31:02)

Maybe.

C (31:02)

Okay, well, that's not what happened here. But they could be doing that.

B (31:05)

You know, they had this like security engineer that like their WhatsApp got hacked by, you know, by a agent or something like that. That was three weeks ago. It just seems like a lot of this coming. I mean, maybe that's just like how their security organization works these days. I don't know. But just seems like a lot of that type of information coming out of there.

C (31:25)

The bots are not. Okay, Britt, I want your take. We should also talk. We should shift clause slightly because guys, Horizon Worlds has been shut down too.

B (31:35)

What's that?

A (31:36)

That meta's thing?

C (31:36)

That was one of Meta's Metaverse products.

A (31:39)

Yeah, but we knew that was happening.

C (31:40)

I saw a great tweet from someone, Kylie Robison, who said it turns out it didn't have legs. Do you guys remember the avatars were like legless, so they got it. That was Kylie's tweet.

D (31:53)

I like the tweet I saw, was just saying that, that it was like meta sets down horizons universe. Both users are very upset.

C (32:03)

Okay, Claws Metaverse, what's your take?

A (32:06)

Well, I'm just going to finish on the clothing and then we can pivot. But I, I want to show this, this chart from open Router. Like you can see the Nvidia effect here. It's just like, you know, Jensen gives his presentation and it's already growing. Like this is open call router usage, right? It's like boom. And this is predictive today, of course,

B (32:26)

also show the rankings down at the very bottom because it's a good, it's a good answer to Sam's question about like, which of these is the big one? Like go all the way to the bottom.

D (32:35)

I thought Kimmy was the biggest model.

A (32:38)

Look at this top apps. Open claw has 800 billion tokens. The next biggest 1 kilo code is 200 billion. The one next Claude code is 100 billion. So it's like literally 8x.

D (32:50)

I am with you. But that's a little unfair on the Claude code side because Claude code's tokens are all going through Claude, right? Like, whereas I don't even know how you make. I mean, I guess you could make Claude code talk to open router, but that's quite a leap, right? Like I, I get it, but like this is literally open clause.

C (33:06)

Guys, can we talk about Janitor for. Oh, that we already talked about Janitor

A (33:10)

we already talked about Janitor. They're number five, I'm telling you. They're spicy chats. Next, you guys. Just saying.

D (33:18)

I'm just saying it makes sense that openclaw would. I mean, again, that's not to criticize openclaw. Like, I'm pro openclaw. I'm just saying, like, obviously Claude code, which does Claude directly. I mean, the fact that it's like almost what, like 10% or actually more than 10% of OpenClaw, given how hard it is to make Claude code talk to open to OpenRouter is like kind of the opposite, I think shows the opposite story.

B (33:41)

Yeah, it depends on which story you're talking about, Sam. Like, you were asking a question about, like, what's the difference between nano claw, open claw, which one is the one? Like, and I think you look at this and it's like, well, there's one major ecosystem developing. There'll probably be more, right? But like, I just don't even know what it is.

D (33:57)

But I guess the question for me is more intellectual, which is like, what even is an ecosystem when everything can be ported so easily? Like, I don't even. You know what I mean? It's just like Internet, right? It's like, yeah, it's like, yes, I get people are using OpenClaw, but if OpenClaw went away tomorrow, it would disrupt nothing, right? It would just be like, fine, you just use nanocl. Like, there's no. There's no cost of switching at all.

B (34:17)

I don't know. I don't think that's how human communities work. Like, you're one of the. I mean, you've articulated perhaps better than anyone that communities of trust matter more than ever in the current Internet. And I think that's partially what's going on here, is we've been trying to do this open cloth thing in a very transparent, very trustworthy way. You know, we've been trying to set this up in a way that respects the community that's growing around it and establishes more trust over time in order to kind of, you know, combat the question that you're talking about. And there's a lot of other ways to do it. Like, if one company had, you know, back to your question at the beginning of the episode, like, what is the OpenClaws found? Like, what are we doing with the OpenClaw Foundation? Like, at the end of the day, open source projects die when one company captures them, right? And so we have to set up a 501C3 that enables it so that no single company captures it, no one can control it. That, you know, we're doing this in a maximally trustworthy way.

D (35:24)

Yes, I think that that's the. Trust is the right word. Which is if you said to me the reason Open Claw is what everyone uses is because everyone else is worried that Nano Claw, the thousand other things that are exactly the same might steal their data. And this one everyone trusts. That's a, I think a valid answer which is like if you're pumping everything through it, then like all that matters is security and trust.

B (35:42)

And you know, the very best open source projects throughout time have tried to do that. You know, like we've been getting great advice from folks at the Mozilla foundation and the Linux foundation and many others. You know, the great open source foundations upon which the Internet has been built have tried, you know, to do their best around trust. And it's, you know, it's a, it's a meandering road, like you've got to do the best that you can. But I think that's, that's the answer here.

C (36:14)

David, at the end is the reason I'm going to use OpenCloud instead of Cloud code.

A (36:18)

Cost.

B (36:19)

Yes, I mean I think that. Which is a good reason. One of the reasons. Right. I've said it a lot on this pod and in many other contexts lately, but you know, there's somewhere on the order of 4, 4 to 6 billion people on the Internet, there are a billion people today using ChatGPT. I don't know, there's some very, very much smaller, ver. Smaller number of people using Claude and the related ecosystem. And this technology should be available to more people in the world. Like, you know, great AI harness plus great model should be available for free to as many people in the world as possible. And so one of the reasons I'm so passionate about this is that I really believe in this idea of personal AI, that you should be able to have an AI that runs on whatever computer you have access to and runs whatever model you can afford to run. And that, that should be available to as many people in the world as possible. Like you shouldn't have to pay to access some form of AI.

D (37:21)

Next week or in two weeks we're hosting in San Francisco a Creators learn how to Use Bots for Real day. Actually two days, including a hackathon. It's going to be really fun and I actually like there. It's, it's, it's a pretty exciting one. So you can, you can evangelize to the creator. It's going to be awesome actually Dave, I think you're going to love it. We have TAs coming, we have a full on hackathon all with big creators and the question is, how do you leverage this shit for real?

B (37:45)

Yeah, I mean one way to think about it is that agents or claws are like the new website, right? And in the late 90s you had to figure out how to make a website for yourself, right? And so a lot of people figured out how to build a website that was about identity or about content. Some people built them for commerce, some people built them for all the myriad reasons. Right. And I think that's like where we're headed. And so it's gonna be like a decade of building agents and so that's great. Like I'm down, come hang.

C (38:15)

There's some alarm bell that just went off for me. I'm interested what people think about this. But in my world I hear from people all the time that information should be free, information wants to be free. You know, that's half the quote. And I obviously agree with that for some information and vehemently disagree for other information that is deeply expensive to produce and would not exist where they're not a business model for the information. So. But this is a tangent, probably not worth going down, but I wonder how we'll see the intelligence conversation play out in that way. And I think we should mention that another headline this week has been OpenAI focusing on enterprise. And following the information's reporting about the code red several months ago in terms of competing with then Gemini, they're sort of entering this another strategic review period where they're going to focus on enterprise and coding and get rid of what they call sidequests, which those among this group will know is a very meta term. One of Mark's favorite things is to talk about side quests that'll be interesting

D (39:31)

for their recruiting because they've recruited. Yes, they've recruited based on the thesis of side quests.

C (39:36)

They have recruited, but they're going to. So what was your take on that? I'm sure you guys saw the headlines.

D (39:41)

I mean I didn't, but I think for me it's like the story of OpenAI is like the business model go round and round of like what it is, how they're going to make money, right? And like we've seen this cycle several times before where it's just like, oh, consumer's hard, enterprise, enterprise hard, this consumer's hard, you know what I mean? So like I'm just like, I don't know, I'm where I'm always on this, which is like they're searching for a business model that's going to make them money. And I think it's very difficult. And like, honestly I do think that if they're on the enterprise side, like, I don't know, having seen the information's reporting, it seems like quite likely that two months from now the story is going to be that Claude is actually making way more money than OpenAI. And I think that's going to be a big blow to their narrative.

C (40:23)

I feel like OpenAI also has a large enterprise business though. I don't like, maybe, but remember, like,

D (40:29)

I mean you're telling your own reporting about those lines. Those lines are your reporting. Right. And those lines say, I know the

C (40:35)

lines are clear, but like it's. The lines could totally cross. The lines could totally not cross being anthropic and open AI.

D (40:41)

I'll just push you on this. You're reporting like over the Last several

C (40:44)

months, 40 reports my team is reporting. Yep.

D (40:50)

Something would have to change in a dramatic fashion for Claude to not cross OpenAI in the next two months. Like something weird would happen.

C (40:58)

Two months would be. Two months would be tough. Yeah, no, I think that's true, but

D (41:02)

on the ARR basis or whatever you guys are calculating.

B (41:04)

So is that the story then?

C (41:06)

I know. So that is definitely one of the stories. And if that happens, that would be a story. And obviously the pivot to Enterprise encoding within OpenAI is because they fear that might happen, by the way, as they're all trying to hit the public markets and raise more money than ever has been raised in the history of mankind at valuation. So the timing isn't opportune. But when I talk to businesses or third parties, they talk about OpenAI having a massive enterprise business and in fact their consumer business gave them an early mover advantage in the enterprise, these people say, because of marketing. And so I'm a little confused.

B (41:47)

Yeah, I mean a huge number of people are using them. But isn't that kind of what they're saying? They're just, they're just saying like we want to reallocate our human resources to focus a little bit more on growing that business faster. Like, isn't that kind of the message?

C (42:02)

Yeah, yeah. I want to know what's going to happen to the device, guys, I want to know what's going to happen to

D (42:06)

the device because like, is that a side quest?

B (42:08)

Did they say they were canceling it?

A (42:10)

Well, that's meant to be a consumer device.

B (42:12)

I haven't seen the stories. Did they say they were canceling it?

C (42:14)

No, I don't think there's been reporting about them canceling it. Neither do I think they will. I also think it's a side quest though. So I think that like if you were really, you know, I don't know, I don't know enough about it, but it's. And obviously there's a lot more reporting to be done, but it, it's really a good old bare knuckle. Oh my God. We haven't talked about my Travis reporting. Guys, that was like. Was that in the last week? It was, but it's like Lyft V Uber times 10,000.

B (42:47)

Yeah, totally.

C (42:48)

How do we all feel about getting in a Travis Kalanick made self driving car?

B (42:52)

I thought we were going to talk about the vibe coding Apple thing.

C (42:56)

Oh, we are, Dave. We are. Okay, moving on to that. So we know we have this explosion in vibe coding apps. And not just apps that are solely dedicated to vibe coding, but vibe coding features, inside apps, Canvas, Snap. Right. All these companies let you do things. And my colleagues reported this week at the Information that those apps are getting stymied in the App Store review process because Apple is claiming it fundamentally changes the nature of what these apps are. So some are getting through. You can obviously access Claude through the App Store, but we hear that a lot of startups are getting caught up in this. What are you guys hearing?

B (43:38)

Oh, man.

A (43:39)

Oh boy.

D (43:39)

My more or less app is stuck in app review.

C (43:42)

Okay, that doesn't count. Dave, what about you?

B (43:44)

I have things to say.

A (43:45)

Yeah, this has been near.

C (43:48)

You have the microphone.

A (43:50)

This is a Dave episode. I like it. The floor is yours again, Dave. No, you should do it. You should do it. You've been fighting this battle.

B (43:57)

I mean, so the information piece is great, Jess. I think we need to adjust the framing a little bit because I think setting the record straight, I mean, we've been living this for like the last six months. We have a company that we're very close with and working on called Bitrig. This is the co founders of actually SwiftUI, which is one of the best Apple frameworks in the Swift world. So a bunch of ex Apple, it's a whole ex Apple team. They're world class engineers. They were part of iOS and iOS engineering for the last like decade. Really good guys. They've been blocked since November.

C (44:40)

Wow. What does the app do?

B (44:41)

It's a vibe coding app, so it was one of the first apps to enable the creation. Like I think of it as an iPhone app that creates iPhone apps. And I was really excited about this idea last summer, right?

C (44:53)

Like, can you say the name again for people who haven't heard?

B (44:55)

It's called Bit Rig.

A (44:56)

Bitrig.

C (44:57)

Bit Rig. Okay.

B (44:58)

Bit Rig. Bit Rig. And, you know, I want to sort of reframe this, though, that it's not really about App Store revenue. Most Bit Rig users, like, want to publish to the App Store. It's also not about protecting xcode. Right. There was some sense, I think, in that article that potentially this is about protecting xcode. I don't think it is. Apple's always allowed, like, React Native and Flutter and like, all these different sort of developer tools. The actual issue, which you guys did mention in the, in the article, is this guideline 2.5.2, which is that apps can't download and execute code that changes functionality. That's always been this kind of gray area in the app guidelines. And like, look, if I worked at Apple, like, I know how seriously they take user trust, right?

D (45:48)

But we just said that's all that matters. That's the only thing that matters for openclaw, all that matters is trust and security.

B (45:54)

Yeah. And so if a Vibe coding app, which was approved by the App Store review process, right? Which, like, when they do the App Store review process, they want to figure out what your app is doing, what data it's accessing. Like, are there any risks, right? Like, apps that are on the phone can access things like HealthKit, right? Like, let's say a Vibe coding app of some kind was able to do this. And after it's submitted and it's on people's phones, it can generate new code that could exfiltrate, like your health kit data. Like, that would be bad, right? That'd be a real problem. And so I think their caution is warranted. I don't think it's malicious here. Probably they've blocked updates, they haven't pulled apps, right. So it seems like they're buying time to figure out the right policy. But, you know, we've got an app that's been blocked since November. We've submitted multiple times. We've worked with the App Store review team. We've gone all up and down the chain. They've all been rejected. And these guys are just trying to run a business, right? And so I think they, plus all the venture capital and people behind all these apps in the ecosystem, just need clarity so that we know where to invest. Right? So what we've done, Bitrigger, we've actually, like, done some of what you guys have talked about in that article, we went server side with it. Now you can generate code on a server in a virtual machine and you can display that in a pop up, right? But even there, like Apple's like, well, you can't, you can't display it in a pop up in the app. You have to display it in Safari so that the user is not confused. And I don't know, I just kind of hope we can take this to a new, a new level of a conversation. I appreciate the article that you guys built out today, Jess. Like, I think this is a complex issue and it's about trust and security around the user's data. But at the same time it's super clear. If you just look at the number of apps that have been submitted to the App Store, it's like that number is going vertical, right? So people are generating apps with Vibe coding tools and agentic engineering tools. And this has become like a really serious part of the ecosystem very, very quickly. You didn't just sit there in the middle of the Nvidia conference yesterday. I mean, this is becoming an enormous part of what we do and how people build apps. And so I think it's becoming, it's like a pressure cooker, like the pot's boiling over. And I don't know, I'm hoping we can have a more sophisticated conversation with Apple about this and figure this out sooner rather than later.

A (48:32)

Our 11 year old even Vibe coded his first app that we are on the brink of trying to submit to the App Store and it's like he's

B (48:38)

built a game to think about.

A (48:40)

Maybe they're just, maybe Apple's just afraid of the influx. Like they don't have enough team and processes and tech and agents to be monitoring like a 10x or a hundredx amount of apps in the app.

C (48:51)

This is like the Apple that told me I couldn't list text on red on my website. So maybe they could just pick some better. You know, I think their battles, I

B (49:00)

think that's it for it. I mean, I heard that OpenAI and some of the bigger apps have been stuck in the queue for a month for changes that are unrelated to Vibe code. It's just random changes unrelated to. It's not Vibe coding.

C (49:12)

They're just like interesting too.

B (49:14)

Like I keep hearing that, that like big apps are stuck for a long time, which, like, that's not good either.

C (49:19)

I totally understand this like triggering new things that Apple has to sort through and figure out and communicate. But if you're Apple Vibe coding is also. It's an opportunity to the point, Davey said of the explosion of apps, but it's also a threat in that you can do more within a chat interface and get more functionality without having to go through apps. Right. I mean, it's both.

B (49:44)

Yeah, I just, I don't think that's the issue here. I really don't like. I think it, I think it comes down to can you, can you build an app, submit it to the app store and then they, they look, you know, they do a lot of stuff. They look at is the code good? Is there anything malicious in it?

C (50:01)

Yeah, no, I know.

B (50:02)

And then once it gets on everybody's phones, can it now generate new code that does a completely new thing inside of its own sandbox? You know, we see this in the open claw ecosystem too, right? Because we're seeing this with all the skills, right, that are being generated. Like we had to build an entire approval process now for skills because the skills that were going into the. That was the thing making open claws unsafe was that people would download a new skill and then there would be instructions in there that would exfiltrate data or do something nefarious. And so it's a very similar problem.

C (50:37)

I actually 100% get it. I also think Apple has a history of saying things are privacy related and then letting them do it and not letting other people do it. Ad targeting, for example. So I think developers and the ecosystem is smart to be watchful of this. And I think it's really early days around these apps and what it means for the ecosystem. But the security issues do make sense. And to your point, I would want them to be policing apps from fundamentally changing and accessing things that they couldn't and that kind of stuff. But yeah, I thought it was a good issue that the team kind of like nuanced Russell to the ground there.

B (51:16)

Yeah, they did a great job. Hopefully I can add and we can add and I think however we can help with this. I think it's a hard. It's a very nuanced conversation that.

C (51:26)

And people do not like talking about Apple publicly. So props to the team for having actual examples. What else, folks? I think we've all. I think we lost Sam. He claims to a technical issue. 10 bucks he's sauna ing. I don't know that for a fact.

A (51:40)

I got booed.

B (51:41)

We are having weird technical issues with Riverside. We've had screen sharing that won't stop all. So hopefully all of our streams are

C (51:49)

okay and this comes across to everybody over Our computers. What's been got? I've got one fun thing from the Lesson household to report, but any fun things from the Morin household before besides Vibe coded apps going to app stores?

B (52:01)

No, it's just all claw all the time.

A (52:04)

No. We had to create our son's first LLC for his Vibe coded app. So that was like a monumental moment.

C (52:09)

Baby's first llc. You must have been so proud.

A (52:12)

We had like a little LLC party, you know, got him a cake.

C (52:15)

It won't be his last Sprint. It won't be his last.

A (52:18)

No. I know he's going to bite and he's going to like overtake me and Dave's number of LLCs by like age 16. I guarantee you.

C (52:24)

It's like, do you celebrate that or going to prom? Is prom still a thing? I don't know.

B (52:28)

I'm pretty proud of him. He made a pretty great iOS game. Like, it's, it's legitimately good. You know, from my, from my product guys, I like it. He's doing it. We will. Once it's, once it's in the App Store, well, we'll tell everyone about it. It can be a, it can be an ad on more or less.

A (52:46)

I think we should all be very cautionary about Gen Alpha right Now, like the 10 to 15 year olds because like their use of AI is growing faster than anyone's. We got intel from Clawcon Austin, the open claw meetup that the best presenters of all the developers of the night were these like 12 to 14 year olds from the Alpha School in Austin, which is like the AI.

B (53:10)

They blew away the adults and it

A (53:13)

was just, it was like unbelievably night and day what these kids presented versus normal. The normie developers, you know, the Gen Z ers and millennials. So I'm, I think there could be like some interesting new cohort or fellowship program or something for like the Gen Alpha AI developer ecosystem. Just, just sit on it.

B (53:33)

I just think it's so cool. You know, when I was a kid, I got, I would go over to my grandfather's house and use his Macintosh plus and it had this app on it called HyperCard which made it possible to build games and it was super fun and it made it, it really gave me the sense that I could create anything on the computer. And it's been an interesting thing. The last, you know, being a parent of kids that are now 10 or 11, there hasn't been a tool that you could just give to the kids and they could create what they want to create on the computer without learning really complex interfaces. If you're trying to like 3D print something. The CAD interfaces are so difficult. And to make software, it's like all the software development environments are really hard to learn. And so it's been kind of magic to watch our 11 year old take OpenClaw and create exactly what was in his mind. And you know, like that to me says maybe the kids are going to be all right. Like they're going to be, you know, they've got a, they've got some really, really fun tools that makes it possible to build things for the first time in a while. And that, that's, that's really to me like a very optimistic take on what's going on right now.

C (54:48)

So in the kids are going to be all right. Camp I finally set up and I'm embarrassed that it's taken me this long. Our board. Do you guys know what bored fun is?

B (54:59)

Yeah, we have one.

A (55:01)

Oh yeah.

C (55:01)

Oh my God. Guys. I let down my 7 year old by adding a burger too late to the burger order in. Chop chop. Thereby adding to our food waste bill, thereby letting down me and my three boys as we were playing this awesome restaurant game. For those of you who don't know, it's. It's a physical game board that's digital with some physical pieces. So it's a combination of digital and physical interaction and it's just so social and so this, We've only played one game so far. Chop chop. And you're making things at a restaurant and you have to collaborate because the tickets come in and then the lunch rush and the dinner rush and my 4 year old's yelling because we only let him chop but we don't let him assemble. And my 7 year old's crazy because he wants to wipe everything down so you don't get the surcharges and it's wildly fun. I cannot. I mean it's wildly fun. And my boys woke up and they're like if we brush our teeth and eat breakfast, can we play chop chop before school? Like they are obsessed.

A (56:07)

That's cool.

C (56:07)

But I really have to get my act together because I didn't see the time was running out on one order and I fired another bacon cheeseburger and I let down the whole family.

A (56:18)

Us millennials are the laggards these days.

B (56:20)

Yeah, just the other one that we got that the next playground. Have you seen this thing?

C (56:27)

No, but tell me.

B (56:28)

It's like a, it's a little cube that you put in front of your TV and Then it has games on the tv, but you have to move your body and jump around. You can play basketball. And somebody had told me that it's the most fun family thing that they've done. And I was like, I don't know, I've seen these things before. But our kids are like fighting over doing this every night.

A (56:51)

They love it and they do it together. And it's because there's, it's like, there's bowl. It's like we, but like way better. And it uses computer vision to like really know exactly where your arms are. You don't need the weird Wii paddles or anything. So you're playing tennis, you're bowling, you're playing basketball, you're like doing all these things. And even our three year old, there's like levels like that are low enough for toddlers to participate. So highly recommend it. It's been a fun new family. Family Tech plus meets physical offline world activity.

C (57:21)

It's great, you know, the ingenuity. Well, with that, friends, I think, I think we covered some good ground and it's time for us to go check on our agents because it's been like a whole several hours.

B (57:32)

So without hr, they might be running amok.

C (57:34)

Sam's master bots is probably at it and. Yeah. So thank you all for listening as always and the feedback. Even if you didn't like this episode, we will forgive you.

B (57:44)

And the selfies.

C (57:45)

And this I love. I mean, I gotta feel like Jensen out there. And with that, I think the ladies will be here bad next week. But not the oh my gosh, we gotta.

A (57:55)

We need to strategize on this.

C (57:57)

This is the point where we figure out our schedule. It's always this. Yeah.

A (58:02)

For anyone listening, we do this very ad hoc.

D (58:05)

No.

A (58:05)

Some people are like, do you guys have like teams of producers? And I'm like, no.

C (58:11)

No one thinks we have teams of producers, but that's not the vibe we're given. But the ladies will be here back week with back here next week maybe with some special guests and we'll see you here. Goodbye.

A (58:24)

Bye guys.

B (58:25)

See you guys later.

A (58:27)

If you enjoyed this show, please leave us a virtual high five by rating it and reviewing it on Apple Podcast, Spotify, YouTube or wherever you get your podcast. Find more information about each episode in the show notes and follow us on social media by searching fororless avemorin essenless lesson. And as for me, I'm Brit. See you guys next time.