You're watching TBPN. Today is Monday, September 21st, 2026. We are live at the TBPN Ultradome, the temple of technology, the fortress of finance, the capital of capital.
How many days till Christmas? TBPN's road to Christmas started weeks ago. We got 94 days until Christmas. Double in double digits.
We're so close. Every day one day closer to Christmas. So get excited. Start shopping for presents. Start picking out that Christmas tree.
Let me tell you about Ramp.com. Time is money. Save both. Easy to use corporate cards, bill pay, accounting, and a whole lot more.
They got musicals. I guess they do musicals now. And I guess we do AI safety now. We'll get into that later. Doing a bit of a show today. Yeah, we got Augustus Doricko coming in with a hundred million.
Tip of the hat for a hundred million. Right.
We have Vinod Khosla. Vinod Khosla's coming on, coming back. Levie breaking his silence on the SaaS rebound, new model drops, the latest on open source and more.
We have Ron from iConnections. They are putting on an incredible show in November that we're going to be at called Tomorrow X. Excited to talk with him about that. And then we'll round it out with none other than Ben Hylak coming in with a new round of funding as well for Raindrop.
Raindrop. How was your week?
Yes. Oh, yeah. Good name for a company. How was your weekend?
It couldn't have been better. It was good. We were off on Friday. We went to Boston and back basically the same day. Sort of a brutal flight back and forth. A little tired, but terrific.
I got completely one-shotted by the new AI video model, Seedance 2.5. Gone are the days of face swaps. I remember when it was called DreamBooth. This paper came out from Google and it allowed you to very realistically swap your face into a photo. Now this is something you can just do with a single prompt, but back then you had to go and take 10 or 20 different photos of yourself at different angles, different lighting. Usually just dug through your camera roll and got them. Then you had to feed that into like a Python script that was either hosted on... what's their notebook? Their hosted IPython notebook, Colab. And then the Colab thing was weird because it was luck of the draw what GPU you got.
So you would like sign up and create an instance and then sometimes it would just be like, you got a 1080 Ti and you'd be like, I can't do what I need to do because I need a lot of memory to do a particular thing. And then sometimes it would be like, you got an A100, brother, like lock in. Like it was just random provisioning and people would be so excited when they got one or the other.
But there was this musician who was like a guitarist who implemented the paper and then basically distilled it down and figured out how to run this. This is a crazy thing, but this guy who's just like a hobbyist machine learning engineer figured out how to get DreamBooth running on like basically commodity gaming hardware. So then you could go and you could take your face, fine-tune it, and then be able to generate new photos in the style of whatever. It was all built on that open source one, not Midjourney, the other image generator, Ahmad Mustafa's Stable Diffusion. It was like Stable Diffusion as the base and then this DreamBooth thing on top. And it was super cool. Because if you actually went through, if you jump through all those hoops, you basically had access to the magic avatar thing like three months before everyone else and you could like send one to a friend and be like, 'Oh, this is cool.' And then it was just completely commodified. It became an app and then it just became available in every chat app very shortly after.
But the AI video has been moving a lot slower. It's very compute intensive obviously since OpenAI had to shut down Sora and it just feels like none of the major American labs are as focused on this. So I got completely addicted to this. Had a lot of fun doing it. I'll walk through everything that we did.
Yeah, there was one point where you sent me like 10 videos back to back of me in various AI scenes and then I was just sending you some like pretty mundane commentary back and then I was like, whoa, okay. On something totally different.
Oh, yeah. Yeah, I was like, 'Oh, sorry. I need to respect your culture and actually respond.'
Yeah. Well, it's because like you no longer need to go get a different app or sign up for anything. I was just texting with Codex remotely. I had a PC that was running this and then I would just be like, I didn't even say take this video. I would just say like, go find some videos and put Jordi in them and it would just do that. And I didn't even have to like upload photos of you. I would just say like, go find some reference photography for Jordi across the internet and it would.
And so far it seems to have made a lot of sense not to focus on video given that Codex and Claude can connect into all these other tools and models and just give you what you want as a user without incurring any of the cost of focusing to make great models or the actual cost of serving them and having to allocate, you know, GPUs.
Yeah. So I think people will be having fun.
Oh, speaking of Codex, let me tell you about Codex. Codex is a powerful workspace for getting work done with AI agents. Whether you're writing code, analyzing data, creating content, creating some of the videos I will show you shortly, or automating business workflows, Codex helps you move projects forward from start to finish.
So the latest turn of AI video models, like they're really, really good, not just for the high fidelity stuff which we'll show you some of, but even going back to like '90s VHS footage, it can edit anything and it doesn't look weird 90% of the time. Seedance 2.5, the big update here is that it replaces the whole person, the whole object as opposed to just a face swap, which just hits way harder. You'll see. And it matches all the subtle details of the footage so things integrate really nicely. Dylan Patel posted a video of him as Drake in this cheetah print video over the weekend and I think a lot of people fell for this.
It tricked everyone. I think it tricked a lot of people because it's super high fidelity. Let's play SemiAnalysis' latest post on X, Dylan Patel as Drake. And what's interesting about this is the video model's really good, but also this Drake video came out a day ago. And so it hasn't saturated the internet. Like a lot of people hadn't seen it.
This is the first... Yeah, I saw this video before I saw the actual Drake music video. And of course, the choreography is fantastic in the underlying video, and that drives the motion.
Really? Yeah, we're probably going to get flagged as it is. But it's so funny to see Dylan walk around and that first shot really sells it. It really looks like him. Later on, he goes into the mirror. He walks up the stairs. This is what the model is really good at, is even though he's very small and he's behind, like that chandelier cuts in front of him. There's no aberrations. This part, his face starts to look a little not like him. There's a little bit of Drake like sort of sneaking back in on the facial structure. But overall, very, very incredible performance and it does a good job whether he's looking away from the camera, whether he's big in the frame or small in the frame. And I can't stress enough how funny it is that so many people were dunking on this saying, 'Wow, this is the top. We're definitely closer to the top than the bottom. Never become the main character. Serious. He's not serious.'
It's like one, Dylan has never had the most serious brand. He likes to have fun. But two, the fact that it was tricking everyone shows where we are.
No, seriously. Like this is exactly what you want from SemiAnalysis. Like you want them to analyze.
We just updated you on video models.
Yeah. And the idea that Dylan, which they use so much AI, they use so much AI on anything. Everything. I even texted Dylan recently. I was like, you should get an actual creative director, maybe use less slop, you know. But the idea that he would just like take 48 hours and a $100,000 budget to go and film this the day after Drake drops the video is crazy. But if he did, like he is like obviously he's in like an analysis business, but he's also an entertainer. He also has to fill the top of funnel newsletter subscribers. Like he has to be front and center. And so it's actually not that crazy. Like even if he did this by himself, I would just be like, respect. Like that's a fun project. But of course, he's using AI. He's using Seedance 2.5. Higgsfield wrapped Seedance 2.5 in a model that sort of, a harness that sort of makes this a little bit easier. You can use that via the API. It's also available on Fal. And there's a couple other ways to interface with it. But it's interesting because it's highly compute intensive. China's obviously way behind on compute. So what's going on with the strategy? But before we get to that, it's really fun. So we got to play some of the more videos that we made. So I had Codex finding famous or funny videos to use as base clips. I would just say, I started with the rap videos. Did a couple of those. I think there's one maybe in the timeline.
Noah in the X chat says no headphones, obviously fake.
Oh, true. Yeah, that's a good point. Yeah, you got to have the headphones to make it believable. Dylan Patel. But yeah, let's see if we can play the video. Did that one make it into the production chat? Yeah, this one, because I saw this originally with Elon Musk and Sam Altman rapping and I think we might just because of swearing.
But what's interesting here is that it does a pretty good job of getting us. But there's moments when I sort of blend into Elon Musk, which is funny because it's not like it's originally an Elon video, but clearly the underlying video that was driving the AI model was the Elon and Sam version. And then our heights kind of change, like I look a little bit more like Elon there. It's very silly. But very, very funny, going to fool a lot of people. But also then I was able to just say like, okay, go make a bunch of tech videos, like just find iconic videos from tech. And it did it and it found like Sweet Baby Ray's. I didn't tell it to go get that video, but it just did it. Which was very cool. Some of them came out really well. Some of them didn't do too well. The one that I liked, I think the most.
And it's a funny dynamic that we noticed where when you're looking at an AI video of yourself, you're thinking that doesn't look like me. And then when other people watch it, they're like, 'Wow, that looks exactly like Jordi.'
Yeah. Yeah. So, let's watch Jordi introduce the MacBook Air as Steve Jobs. I don't think actually, Apple kind of bans everything. So, we'll see. We'll see if this one gets flagged, but we might be cooked on this particular one. But, let's play. I said video is solved.
Yeah, let's roll the dice on this one. Video is solved. Jordi Hayes introducing the MacBook Air. It's already got 350 likes on X. Let's play it. Boom. So, it didn't change the voice. So, you still hear Steve Jobs's voice, which is also funny about the Dylan thing because it's Drake's voice. So, it's like, what did they think happened? He's like lip-syncing to it or something. That didn't make any sense. But what's really cool about this is that it changes Jordi, the small Jordi and the big Jordi. It changes both of them in the same frame. And then even there, you only see a little bit of Steve Jobs' mouth. And it was able to change that to Jordi. So it can represent what's going on outside of the frame. And historically, that's been very difficult for AI models. They want the face in the center of the frame. They want to be able to crop it and think reasonably about it. But it did a good job there. Jordi as Jensen taking the GPUs out of the box for this one.
Yeah, it's very funny when you use a different person with a different stature, different musculature, like it just comes across differently. But very funny clip, iconic when Jensen announced, was this the H100 announcement, the A100 announcement? It was pretty early, but for a while.
This is still such a funny gag that Jensen actually did this. I love it.
He's really grunting. I mean, that's a heavy rack.
Got here. It's the world's largest graphics card.
Such a funny bit that he just did this in a kitchen. It's so good.
Something I got to show you. I got something I got to show you. Jensen's such a master performer. We saw that on stage at Salesforce Agentforce last week. The last one that I did, and before we pull up the Jordi as Steve Ballmer, which is particularly hilarious. Let me tell you about Console.com. Console builds agents that automate 70% of IT, HR, and finance support, giving employees instant resolution for access requests and password resets.
This one I specifically told it like keep the sweaty shirt. And so I think it did more of a face replacement. So the face is sort of jumping around, but still very funny. Key to .net, the key to industry transformation, the key to success is developers, developers, developers, developers, developers, developers, developers, developers, developers, developers, developers, developers, developers, developers, developers.
It really got like the level of sweat on your face pretty correct, but the face is jumping around. Not quite dialed. I think you'll have to perform that one IRL.
And I should run that build at some point.
You should bulk to get to that level for sure. For sure.
So yeah, I mean downstream of this obviously there's going to be the 'it's over for traditional filmmakers' blah blah blah. Then obviously a whole bunch of blended work that uses the right tool for the job. Some productions will use this where it helps in urgent scenarios. I think some of this stuff is already actually being used in Hollywood specifically for reshoots. About 10 years ago, Henry Cavill was in Superman and he left Superman. They wrapped production. He went to film Mission Impossible. In Mission Impossible, he has a mustache. So, he grew out a real mustache, was filming Mission Impossible, and they were like, 'We didn't get the shot for Superman. You got to come back and film another scene as Superman.' And he was like, 'I can't shave my mustache.' And they were like, 'Okay, we'll fix it in post.' And in the newsletter, there's a photo here that we can pull up of how it came out. Because they had to remove and reconstruct his lip and his mouth using 2017 technology, and it was not good. It was very, very rough. Probably very tedious, very expensive.
Isn't a mustache like the easiest thing to apply?
Oh, like fake. Like you should have just shaved it and then put on a fake mustache. Yeah.
I don't know. I think Mission Impossible, they wanted the real thing.
So yeah, the Superman on the right is Henry Cavill with a mustache that has been removed in post and it was widely critiqued by the VFX community. So you can imagine for stuff like that, these tools will creep in very quickly. For most things, it's going to be, okay, well, if you needed someone to drive the video, why don't you just have the actor do that? Maybe for stunts and stuff you'd do it, but for most performances I think you'll still be doing it the old way, at least for the time being.
What about some of the other more action-oriented?
Yes. So this was a fun one. I had to go and pull Johnny Knoxville and Steve-O clips throughout their filmmaking and TV generational run. Their entire run. Make about 20 of them and then put it together into a trailer that we can watch now. And I really enjoyed this. We got to do that in the office. I think Tyler was thinking about getting that.
Was terrifying getting in a cactus. I don't know. I've never really been poked by a cactus. The hands on the bikes. I don't know how the guy on the left wins. Of course, a lot of Steve-O in that one. A lot of Steve-O here, too.
But I think these work so well because a lot of the underlying footage looks like AI. It looks like this. The prompts. This looks like an AI prompt, like drive a jet ski in a pool in a luxury house, you know? This looks like...
This is real though. This is real. We actually shot this.
Yeah, this is real. No, the only thing that's different about this is the suits. We replaced the characters. But the underlying footage is all real. From Johnny Knoxville and crew, the fire hose rodeo. This one that looks fun.
The old alligator in the house.
Like the old alligator in the house trick. I think they chilled that guy down, so he was real slow.
And then the bull run slammed.
I didn't know rams were so aggressive. The ram just slammed that guy.
I know, but I think of them. This looks like an AI prompt, too. Boxing match on stilts, but replaces the suits. The gag chair that falls apart. That's a classic. All-time favorite classic.
Then I come up and, hey, don't steal that. The old air horn while you're golfing. Classic. No, this is a good one. Going through the window. Soup. Just asking someone, 'Hey, can you go grab the six bowls of soup on a tray and bring it in?' Okay, that's just pure Johnny Knoxville. So, yeah, sometimes you get like a blended half the main character, half who you tried to replace. Sometimes it's pretty good. There's a little bit of us in there, but very fun to be able to just do this via text. Like you don't even, I mean, I'm technically interacting with an API, but I didn't write a line of code.
Yeah. All on my phone. So very, very fun. I think these things will sort of burn through pretty quickly. Like everyone will be like, oh yeah, like that. Like that's the capability. Everyone did their version of the meme and then it becomes more of a tool, melts into the background.
Yeah. You already saw this with all the Yung Lean music videos.
Yeah. Kind of made the same thing and now it's played out. But a very impressive demo. Tip of the hat.
Tip of the hat to Yung Lean.
But a very impressive demo because it's able to replace multiple people and do it very consistently even through like some really aggressive motion. So, I thought that was pretty cool. So the whole pipeline here was I was running Codex on my PC, interacting with that from my phone, the ChatGPT app, and then I had Codex go and generate reference images with just ChatGPT images because it's pretty good at that, and then use those to swap through Seedance, which was actually rendered via Higgsfield. Higgsfield, amazing in terms of just delivering the assets like as an API, extremely confusing as a product because they basically have two ways to give them money. You can load up API credits and then you can buy a plan. And so at one point I ran out of API credits and it was like, time to upgrade to the plan. And I was like, okay, cool. I'm having a good time. I'll do more of this. Click the button. I'm like, okay, yeah, you loaded up a couple hundred bucks in here with like the ultimate plan or whatever. And then it's like, those credits are not available in the API. And I was like, great, thanks. And then it was like, okay, time to add some credits to your plan. I was like, okay, this is what I need to do. I add credits to the plan. It's like, those credits are on your plan. Those aren't API credits. Those are plan credits. And so I have like so much random money stuck in various pieces of API.
Well, and I think that's been a massive complaint. Higgsfield has faced a bunch of, I think, various allegations over the last year around just like... I mean, the website looks like a gambling company. It'll be like countdown 10 minutes to claim 75% discount on your next credit. It's like it's totally optimized for like brain rot as you go through the flow being like, oh, you got to buy a plan.
I had seen stuff of like you can't, you know, not being able to actually cancel your plan. So now that you have a plan that you can't use for anything, it's over for you.
It's over for you. You're a customer for life.
But the products, great entertainment.
Great entertainment. And I do think, yeah, I think I might try Fal again as more of like a cleaner API that's a little more intelligible. But yeah, I think once you set up the bindings and get them all dialed in in Codex, I think...
Ellie says, 'Looking mighty sus and mysterious today, chaps.'
Well, this is the era that we're in.
This is the era that we're in. You got to take a safety era.
You got to take a hat seriously.
John, once a week, John's like...
I think fedoras are going to come back. I think fedoras are going to... There's nothing fundamentally wrong with the hat.
I do think the hat can work if it's styled properly. I think it goes better with a suit. I think it goes better with some sunglasses, a jacket.
No. I mean, like obviously the fedora is often maligned as not particularly stylish, not particularly modern. But if you go back to old-school James Bond, there was a way to make a fedora look very stylish. Now we are wearing relatively cheap fedoras. I think you drop a couple thousand dollars on a Borsalino weave fedora, it's a different conversation.
Different story. No, I think it can be done. I think it can be done properly, but it requires a discerning eye, some taste. Taste is the new moat in AI. The fedora will be a piece of that story.
Anyway, let me tell you about Public Investing for those that take it seriously. They got stocks, options, bonds, crypto, treasuries and more with great customer service.
So my questions about Seedance 2.5. I was like, how is China winning? How is China winning in this? Like they have won the compute war. Is it that simple? And then also, why aren't they open sourcing it? They love open sourcing everything. They're like, 'We love open sourcing except when we're on the frontier.'
And there's a weird dynamic there. So, first off, did some digging, got some answers. Let's dig into it. At a high level, intellectual property rules are somewhat flexible over there. That's true. But the models are insanely compute intensive. Like it is, even if you're like, I don't care about Hollywood IP, come sue me in China, good luck, you're still like, there's still a finite amount of compute. And so why aren't you on the RSI trajectory? Why aren't you working on coding models? Like why don't you just take that compute and put it on the next GLM run or the next Kimi run, right? Well, there are reasons for that. First is that ByteDance has a more natural flywheel here than the leaders in America. So, Anthropic and OpenAI do not have good surfaces to deliver video, AI or not video. Like, OpenAI tried with Sora, got a decent amount of people to use it, but it was not the obvious flywheel that ByteDance has. And so, over in China, ByteDance is a big company leading, and they have TikTok, Douyin, CapCut, ads, e-commerce, recommendation systems, tons of immediate benefits to the core business without needing to pull together a new demand surface for video.
This sort of exists in America with Meta and Google, right? But AI slop is deeply unpopular in America. Although I would debate whether or not this is sloppier than some of the videos we showed you. But both companies, DeepMind and Mistral, they are on the RSI trajectory. They're focusing on agentic workflows, personal agents, knowledge retrieval, coding, recursive self-improvement, AGI. It feels like they are more AGI-pilled than ByteDance, which is like, look, we got a good thing going with Douyin and TikTok. We got a lot of video space. Let's dominate there because we have an opportunity. And also like the AI, the IP infringement thing is not actually dead to rights. They haven't necessarily been caught in any major way. It's just that when the model comes out, people are like, Tom Cruise, and it's like perfectly Tom Cruise. And so it's like, what's going on here? So some of that could be the outputs and the harness that actually reviews outputs, because there is an argument on fair use for training to include everything, especially when you think about TikTok. So people go to TikTok and they upload clips from movies. Some of those are fair use, some of those are transformational. Like you can actually make a vibe reel from Top Gun that is transformational and passes the fair use test, which is that it's not a replacement. So you're not like, why would I watch Top Gun when I could watch 30 seconds of Top Gun edited like a crazy hyperlapse edit. So it's not competitive in that way. It is transformative and it provides some commentary layer on top. So that is actually okay. You can use clips from movies if you're commenting on the lighting and you're trying to explain the lighting. All that means is that some of that data goes onto TikTok and lives there forever and Hollywood can't say anything about that. Same thing with YouTube. Same thing with Instagram. There are plenty of film criticism channels that post fairly long clips from movies completely acceptably and they don't even need to share revenue because it is transformational. It is open. It is fair use. And so they have Hollywood IP in the training data. They can train on it. Now there's a question about what the output should be. You still own your likeness. And so even if you can perfectly recreate something, you might still get sued.
Google has a, we tested this out. Google has a music generation service and if you go and say like, I want a Madonna song. Clearly YouTube has all the Madonna songs in the catalog. They can train on them, but it will say, 'I can't exactly make a Madonna song. I can make a pop song that's generic. Here you go.' And that doesn't pull enough from Madonna to get sued, at least in theory. We'll see how it plays out. And then also like a lot of people just upload full movies to TikTok and stuff and they'll just be like, I uploaded all of Lord of the Rings and even though that probably gets taken down, it goes through ByteDance servers, might get saved somewhere, might get like soft deleted, like it doesn't exist on, you can't watch it on TikTok, but it still exists on the server. So they have a lot of data to train all this. But they're not open sourcing, they're not doing open weights. They're certainly not open sourcing the training dataset. And it's a little bit of a narrative violation from the default Chinese AI strategy.
There are a few reasons for this. First, no one wants to open source when they're in the lead. We've just seen this time and time again. And second, with coding models, open weights are a play to build a Chinese tech stack. And so you open source the coding model, build the developer ecosystem, and then maybe that helps Huawei stay competitive. Maybe that reduces the amount of dominance of the American technology stack. And so there's a little bit of like public incentive to build up the indigenous supply chain that comes from that, less so in video where they are vertically integrated and have everything down to the distribution surface. So, in general, if you are a challenger that needs an ecosystem, you want to open up. We saw this with the early Meta stuff. And then if you're an incumbent and you have distribution, you want to close down. So there's some interesting knock-on effects mainly around a speculative future value of video models and training humanoid robots. Like, China also seems to be doing really well in humanoids. If AI video becomes really important in sim-to-real translation and training humanoid robots with simulated video, then this could be like part of the bigger picture. But I think that's further out. I think for the next month we're just going to be talking about memes and funny videos. Anyway.
Let me tell you about the New York Stock Exchange. Want to change the world? Raise capital at the New York Stock Exchange. Just do it.
Do it. Meta on a tear. Meta's up 27% this month according to Shiel Monot. Almost entirely thanks to Muse. This makes sense, right?
Is that almost entirely to Muse?
I mean, the SemiAnalysis... basically when you have a company that's spending hundreds of billions of dollars on something without much to show for it and then you have something to show for it.
Then the market's going to react positively. Yeah. They're saying, 'So, you're saying there's a fighting chance here?'
Totally. And the feedback on the product is incredible. There's a lot of investors that have tried it and they're like, 'Hey, this is a good product.'
Yeah. And interestingly, like they are in the lead in a subcategory where they have never been the lead before. Like Google had Nano Banana, which like there was a moment where it was like, oh, Nano Banana is the best. Obviously Anthropic had best coding models for a while, like crowns have changed, but Meta's never really held the best model in knowledge retrieval or when they did, Vibes, that was not the best version of that. But this is at this moment in time the best version of what they have. And it's building off of a model that's not on the frontier, but able to deliver an experience that puts them in first place for this subcategory, which is exciting.
And can immediately be incredibly integrated across everything else that they do, shopping, messaging, all these different things.
Shopify and Muse announced a partnership this morning, counterpositioning a little bit against Amazon. Amazon came down with the banhammer. Bucco Capital says, 'Amazon cuts off Muse. While I am bullish Meta and Muse, I think many people are overlooking the digital knife fight that's about to occur. Nobody wants to get commoditized or layered here. Let the games begin.' And Amazon says it has cut off Meta's new Muse personal AI agent from shopping on Amazon.com on behalf of customers after attempting unsuccessfully to get the Facebook parent company to voluntarily exclude the e-commerce site from the experience. The problem Amazon says is that it never agreed to any of it. Meta didn't tell Amazon that Muse would access its store. The agent itself doesn't identify itself when it browses and it appears to capture and store customer credentials which the company says create privacy and security risk. As of Sunday night, people trying to use Muse to shop on Amazon were seeing the popup, 'Continued access by an unauthorized AI agent violates Amazon's conditions of use,' to which our customers have agreed.
Interesting. I wonder how Walmart will respond because a lot... Walmart, the overlap is higher than I think most people think. So, I think Walmart had always been more friendly with chat apps, specifically ChatGPT.
And that's the kind of thing that you do when you're not the leader in the space, you're more willing to experiment.
I think they wound up rolling their own. I think they have Sparky and Amazon has Rufus. And so there's two dog-named AI agents that do the shopping. And then most recently...
Tyler's been dailying Rufus, by the way.
Yeah. Just for everything.
Just, yeah, just psychology.
Yeah. Getting close to solving another millennium.
Yeah. I know Amazon wound up finding sort of an equilibrium with OpenAI by plugging in their ad network into ChatGPT instead of letting ChatGPT check out. This is the whole Eric Seufert 'agent commerce is a mirage.' Instead, Amazon is vending in their ads. So ChatGPT brings the data and says, 'Okay, this person's looking for this thing. Do you want to buy an ad on it?' And so there's the transaction value. But I don't know how it'll play out.
Yeah. I saw a couple seemingly smart investors that are using Muse and excited about it say, 'Why is Amazon banning Muse? This is so shortsighted.' And then I actually saw in two instances they deleted because of course they immediately were getting comments about Amazon's ad business which is obviously what they're looking to protect. You can't serve ads to Muse agents or Instinct or Chat or any of these other things.
And Amazon's ad business is actually dramatically bigger than both Anthropic and OpenAI over the last 12 months. Yeah. So they've done 76 billion in the last 12 months through Q2 of 2026.
So obviously they're going to protect this.
Obviously they're going to protect this. I think consumers would love to be able to check out with other services.
It is pretty gnarly to navigate.
Yeah. Amazon.com, as amazing as the service is overall.
Describing your experience as gnarly.
No, it's just being a UX researcher at Amazon, they're calling it gnarly.
Yeah. Yeah. Yeah. But you can see that there's...
And again, Peter, who effectively is the founder of Muse because obviously Muse is based entirely on OpenClaw and they built an amazing sort of effectively almost like...
And I think Instinct too is some OpenClaw lineage. Is that correct?
Yeah. Peter says the beauty of running a claw yourself, they cannot block you. So, oh, they're going to block me when I say claw, open 10,000 Amazon windows on my desktop. Open up 5,000 Chrome tabs of Amazon. So, the souvenir, you can get banned even if you're running it locally. This has happened to me. This is not a full solution, but he is right that there will be more sovereignty and the users might win. Like that's what we were talking to John Quinn about was that Amazon versus Perplexity Comet. The ruling was that Amazon can't block it if it's running on the person's computer because it's like, yes, I told the agent to open the Chrome tab, but it's my Chrome tab on my computer. And so you shouldn't be able to block that because it's not a third-party company. It's not corporate warfare. You're coming to me, the consumer. The customer is always right. I can see that being the ultimate outcome. I do think that's a big... I mean, you basically, I think Amazon's ad business, at least their search ad business, is something that eventually you might need to underwrite like cable television where it's like...
You're basically saying that there's a terminal value here. People are going to use their assistants to shop. But some amount of people are just going to keep going to Amazon.com. Right now, Amazon has made it quite easy to buy things on Amazon.com. It's not like it's that much easier with an assistant, but something I found is it's actually really nice having a single interface for all the things that you do, right? So I like going to one place.
Agent, push the order burrito button.
That's really the future we're in. The Souvenirator has had some of the best commentary on this which actually dates back to, he says, 'The knife fight already took place. It ended a year ago when I wrote the first entry in my agentic commerce is a mirage series. Amazon began blocking agents last July. It's currently suing Perplexity over autonomous shopping with Comet. Amazon won't capitulate. It doesn't have to. Alexa for shopping generated 12 billion in incremental revenue.' Every time I hear Rufus or Sparky, Walmart Sparky has generated similarly significant commercial traction for the company. Agent commerce is already a reality but it takes place on-site, not through independent agents. ChatGPT launched ads in February 2026 and shortly thereafter shuttered instant checkout. The affiliate model is economically suboptimal relative to advertising. ChatGPT has already demonstrated that.
Yep. Advertising is the dominant business model. Yeah, it'll be interesting. I mean, when the OpenClaw stuff came out, I was like, it feels like Napster. It feels like the Napster moment. Like, yeah, you can go check out on an e-commerce website, but like nobody wants you to do that that way. And like you're going to have to write some API binding and then like scrape the front end and then rewrite it every time you go and check out because they're constantly going to be fighting you and you are sort of like playing this like game of cat and mouse which you can play. And for certain e-commerce sites they're going to be very open to that. Other ones are going to be a huge hassle to deal with.
Yeah. And the question is can agents monetize with advertising? I saw someone, a pretty viral post, somebody frustrated with Instinct. Instinct coming and they did basically a blast that says, 'Hey, you have a bunch more invites. You can invite a bunch of people.' And the user was pretty frustrated with that because they're like, 'Hey, you work for me. Why are you advertising?' And so even though they're not paying for the product and they're incurring a lot of costs, the product's free to them, they're still mad that they feel like the agent is doing something that's purely in the agent's benefit, not in their...
Sure. Sure. Yeah. I wonder if we'll live in a world where, you know how ChatGPT has like those recommended agentic actions, if like one of those could be like sponsored, like hey, like want me to go shopping for you right now, like...
Well, yeah, no, and that's important because companies will happily pay for incremental demand, but the issue with agents is if you're saying go to Amazon and buy this thing, the user already wanted to go to. But if I'm an industry coalition of surfboard wax makers, I might choose to send you a notification when you open up ChatGPT that says like, 'Hey, want to go surfing? Want to take it to the next level?' Isn't that what happened?
Basically, you got one recommendation.
ChatGPT recommends... what was it?
Various agentic workflows for me.
Yes, I sent it earlier. Just says one of them was...
It recommended setting up an agent to give me one realistic next step for my surfing each week.
Yep. That's great. It's like every...
Problem is I know exactly what I need to do.
Like, this is actually...
Raj Verma called out something interesting which is that Threads started free-falling in the charts right as Muse started growing.
Yeah. Today on Instagram there's a hero story that's a full-bleed ad for Muse. So they're pushing it. They're pushing it. And yeah, they're probably saying like, 'Let's funnel it over to Muse.' And it makes sense. The monetization on Muse I would expect to be way higher than Threads. Threads, you get some display ads. It's text. That surface has never monetized as well as Instagram. So the best case was like you get something half as valuable as Instagram, a quarter, tenth as valuable, whereas Muse is like a new surface area where there's a whole bunch of different monetization opportunities and they already have the whole small business community. So I think it makes a ton of sense to funnel the attention over there.
Let me tell you about MongoDB. What's the only thing faster than the AI market? Your business on MongoDB. Don't just build AI. Own the data platform that powers it.
Now going over to the less good news in the agentic future that we're all living in. It feels like hallucinations are coming back. There was a moment where it was like if you hit GPT-6 Pro with a query on some history or you basically want a Wikipedia page that's generated or a history report, pull some data, you could rely on it that it was not totally making things up, directionally correct, bunch of really high quality links, pretty valuable. And so hallucinations this year have kind of gone away. You still get slop if you're doing something on the frontier like trying to design some system in code or using Blender or even AI video. You'll see little things here and there, but for knowledge retrieval, the hallucination problem kind of went away, but it's coming back in full force. David here said Instinct made up his middle name and he can't board his flight because he tried to use his agent to book a flight and it just made up David Bomb Chang Choy, which I guess is not his middle name. And I've had that before where...
Well, one, we know why hallucinations would be happening, which is like if you're serving a product for free at scale, you have to figure out a way to do it really cheaply. And so they're certainly using older models and trying to find...
Someone figured out the model. Was it, I think it's Kimi running OpenClaw, right? But it's hosted in the cloud and so you get an OpenClaw instance cloud-hosted.
That said, I've been in situations where my name was misspelled on flight details and this is pretty easy to resolve. So hopefully...
Oh, you think it's not that big of a deal because like you can just get on the phone.
You can just talk to them and they'll be like, 'Here's my ID. I made a mistake before, fix it,' and they'll give you a new boarding pass.
Yeah. Yeah. Yeah. They might kick you out of security lines. So, if you're running really late, as I like to do, you might struggle to get on the plane. You might miss your flight, but that's okay. You got to miss a couple.
This was interesting from former Meta employee Nikita Bier. He says, 'How important is AI to Meta? Here's one easy tell. One of the launch features of the Muse app was auto-negotiating for stuff on Facebook Marketplace. The poor product manager who owns Marketplace knows that millions of bots lowballing sellers will eventually kill the product. But this was escalated to Zuck by the Muse team and he said, 'Okay, we'll poison Marketplace just a little.'
Wait, why will that kill the product?
I mean, think about the experience if you tell your bot like, 'Hey, I'm interested in these types of items. Go anytime one of them appears, autonegotiate.'
Yeah, but wouldn't someone have a bot on the other side dealing with the negotiation?
You can do the outbound, too.
Message everyone on Facebook.
Yes. I got a Dodge Hellcat. If you know, you know. I got one pair of keys and I saw you were post. You liked a post from the Dodge Hellcat community. Would you be interested in purchasing this? I'll need cash.
Don't ask questions. Let me tell you about CrowdStrike. Your business is AI. Their business is securing it. CrowdStrike secures AI and stops breaches. We got Augustus Doricko back in the Ultra Dome, back in the waiting room. He's coming in hot off the Series B. How you doing, gentlemen?
I'm blessed. It's a pleasure to be chatting again.
Congratulations, you align all this news. Was that strategic? Was this like a... You're like, I need to pump the Series B, so I'll get engaged to get an extra press cycle going.
Oh, yeah. They were close together, I guess. Uh, that was coincidence.
They both embargoed same weekend maybe. Yeah, that obviously. Congratulations. We're very happy for you on the engagement, but also the Series B. How are things going? What's new in your world? How much rain are you making these days?
Yeah. Well, first of all, the $100 million that we raised, I'm super grateful to have done that with some great partners.
Thank goodness we keep... You're going to need a bigger gong the next round.
We've been with... I feel like we've been riding with you for every round to date.
And uh, yeah, time to upgrade for us.
No, you were there. You were there early. I was wondering whether a $100 million bag was big enough for the size gong, but for sure.
The reality is, it's a drop in the bucket relative to what the Chinese Meteorological Administration is spending.
Oh, yeah. What are they? They understand. Say again. Yeah. What are they spending? Order of magnitude.
So, they're spending hundreds of millions of dollars a year on their weather modification program because they understand that water and the weather is a strategic capability and resource. We talk about critical minerals. We talk about oil and natural gas. Water is this fundamental input to our industrial economy, to healthy agriculture and food. And if we don't produce more, and cloud seeding is the foremost option to produce more, the Colorado River, the Great Salt Lake, the bodies of water across the United States are going to be impaired to the point of mass evacuation. So, we need to make more for the sake of the strategic health of the country and then also for our allies overseas that need more water like Jordan in the Middle East, like Morocco and so on.
Did you get saved by data center backlash? Because there was some backlash brewing to your crazy scheme to make it rain and then I feel like it's been real quiet since the data centers started sucking all the air out of the discourse.
Well, the data centers have been sucking water out of our reservoirs.
How real is that from your perspective? You're the water expert. Talk to data center people. They'll be like, 'It's closed loop. It's like a car wash. Yeah, we use water, but we recycle it at the same time.' You know, water usage, it's low, but everyone's saying we're going to build a thousand times more data centers, so at a certain point might be significant. What's your take on it?
Well, the reality is yes, American agriculture is using far more water, orders of magnitude more than data centers. So does municipal use, the showers that we take, the dishes that we wash, orders of magnitude more water than data centers. But data centers do consume water. And when you have as great a shortage of supply as we do in the American West, do you want to give water to computer or to your shower and your...
Probably computer. Probably. Weekend all weekend generating AI videos. But if it's generating AI videos and playing video games, I'll skip the shower. I'll walk in. But John, Jordi, you can have AI slop videos in the shower if you just make via cloud seeding.
I can have my cake and eat it too with cloud seeding.
Is there actually any near-term flywheel there where cloud seeding can alleviate water usage from data centers or is that like further out?
Rainmaker already has one deal closed with a hyperscaler and we're in negotiation and engineering feasibility with multiple others. Offsetting their consumption and riding the wave of hyperscalers to produce more water for the American interior. That's a spectacular positive externality that those businesses can provide. And so I look forward to that. Maybe not as much as I do the ski resorts that we're negotiating with. But keep an eye out for the cantons of Switzerland and your favorite resorts around there in this upcoming winter.
That's cool. Yeah, talk more about that as personally as a skier. I'm quite fascinated. But what are those deployments going to look like? Like, is that basically you have a team that's set up at an individual mountain and then you're basically... I mean, these are businesses that have basically dealt with making artificial snow for such a long period of time. I'm sure they took like very naturally to the concept, but what are these deployments going to actually look like?
So they haven't taken supernaturally to it initially because cloud seeding and weather modification has mostly been full of fugazi and snake oil for the last hundred years. Right? There are many, many people in the last couple decades that have claimed to be able to modify the weather but they can't physically measure it and prove it in real time. If you can't do that then why would somebody bother paying you? It's only because of the recent results that we had in Alaska, in Oregon, Idaho, Utah, where we're making hundreds of millions of gallons of new water that they've taken an interest. And we wouldn't be able to do that if not for our efforts to build the best atmospheric science lab in the world. The weather is something that's historically underinvested in by basically everyone besides commodities desks. We need to put way more sensors into the field. We need to build far more sophisticated models in order to understand our interventions and then actually harvest the resources available to us in the sustainable and renewable reservoir that is the atmosphere, be it for ski resorts or farms or data centers.
Are you going to stay with drone delivery for the foreseeable future or is there a howitzer in the pipeline? Because I've seen though some ski resorts actually have howitzers or cannons that will just launch conventional munitions to cause avalanches to clear avalanches for safety. But they already have some of the equipment and I imagine with the right round you fire the right chemical and disperse it at the right... it doesn't seem that crazy to me, might be cheaper, but I'm sure you've looked at it. What's the decision tree for you?
At the end of the day what I care about is delivering on the effect that our customer wants, right? Do you want more snow? What's the best way to get it to you? If it's a rocket or a howitzer or me skydiving and halo jumping through a cloud with silver iodide in my hands, like whichever way is best works. We're taking a really big and deliberate bet on autonomy. Drones, the ones that Rainmaker's built are unparalleled in their capability to fly in these severe atmospheric conditions. Much more wide a flight envelope than most manned aircraft. But in the limit, what we get out of autonomy is more flights, more cost-effective observations, and significantly lower cost per volume of water or snow delivered to our customers. So, we're going to keep using our drones and make them increasingly autonomous, although we'll probably do some research with planes in the not too distant future.
How does El Nino affect your sales cycle? Like, is that going to put you out of business? This is going to be a rough couple months that I just... but I just imagine there's some players that are like, I think we're going to be good this year. Let's check back. No, no, it's kind of a cyclical business, right? Shift your focus to other parts of the US.
It's a... you know, I think Tom Morgan commented recently that building Rainmaker is like playing Diablo on insanity mode for startups. Not only is it a seasonal business because we're making snow in the winter, but also it is annually cyclical because of El Nino, La Nina. Some regions are going to get a lot more natural snow, but other regions will get a lot less and Rainmaker will be working harder there. That said, even in the regions around this, Pacific Northwest, northern Rockies that are getting more snow, they're so critically depleted that they need excess snow melt relative to the baseline year in order to refill Lake Powell, Lake Mead, the Great Salt Lake, etc. So we're geographically hedged in order to accommodate this, not just within the United States, but across the world. That is going to scale up the extent to which we're hedged across multiple different climates simultaneously.
Mhm. What does the team look like? Where are you... have you expanded beyond the Gundo? I know you have like four deployed operating bases, but what's the footprint of the company right now?
Well, you asked first what does the team look like? I first would say Drippy. Oh, show you a series of Hermes ties that we've decided to series with.
This is a nice collection of different ties that have clouds on them. And here you go. Are you donning ties more or less often than you were last year when I saw you on local TV every week?
More often. And you know the Gundo thing, it's great. I love it. I am a Gundo supremacist. Absolutely. And we couldn't have built Rainmaker if not for that. But we want to serve water for everybody. We want to serve water to the federal government, to nations around the world. And in order to do so, we need to collaborate with the best national labs and universities across the country. Being the mulleted guy in Yeezy is not necessarily the outfit and persona that you want to don when you're talking about those research scopes of work. So I'm wearing a lot more Hermes. I am wearing a little bit less Yeezy. And that'll probably persist for a while.
Last question, last question for me. As some of these open problems, Millennium Prize problems have been solved, a lot of people outside of San Francisco are like, 'Cool, the nerd solved another math problem.' Like, you know, they're not as captivated by it. But something that I've been thinking about is like for every one of those sort of high-profile open problems, there's tons of businesses that have access to the same models or models with close to the same capabilities that are just applying them to their industries. And I feel like for you and in your category, it just feels like having intelligence on demand that you can apply to these problems that like hasn't had 10,000 PhDs working on it. You talked about how hedge funds have paid a lot of attention to the weather because of commodities trading, but are you feeling like an acceleration in your team's abilities because of new models or are you not feeling it yet?
To give credit where credit is due, the nerds in San Francisco have done a spectacular thing. And tip of the fedora perhaps. Yeah. Yeah. Yeah. Tip of the old fedora. We're worth taking a point too, right? Like Richard Feynman was a spectacular scientist. Oppenheimer, spectacular scientist. We can do great in frontier science without being total nerds. Just a reminder to the San Francisco audience listening. But again, you can see the rate of hypothesis generation and experimentation accelerating at Rainmaker because of the models available. Astra has been a spectacular lever for Rainmaker and we're going to continue using it. And then also like developing better and better models with the informed participation of the LLMs critiquing the output, like that is something that is actually genuinely accretive to our understanding of the atmosphere and it's going to become all the more so in time.
It's amazing. Awesome. Well, congratulations on the round. Congratulations on the engagement and safe travels. We'll talk to you soon.
Great progress, dude. Great to see you.
Let me tell you about Figma. Agents meet the canvas. Your AI agents can now create and modify your Figma files with design system context. We have Vinod Khosla coming in the TBPN Ultradome in just a minute. First, I wanted to... we're going to talk to him about personal agents and I wanted to share this hot take from Alex Kerr who says, 'I think Silicon Valley is making a fundamental mistake with personal agents, optimizing entirely for the outcome and forgetting that for a lot of things, the process is a very core part of the experience. Travel is the obvious answer. People almost always never want book me a trip. They want to browse, compare, daydream, change their mind, send options to friends, and eventually book.' I think you can actually do all that with agents, but he says the average person probably has far fewer recurring tasks worth delegating than the AI industry seems to think. That's why personal agents feel more like a feature that gets absorbed into existing products than a standalone category. Well, Vinod Khosla has seen both develop over his career. He's the perfect person to talk to about this and so we're very pleased to be joined by Vinod Khosla, founder of Khosla Ventures. Vinod, how are you doing?
Welcome to the show. Thank you so much for joining us. I would love to hear about how you're thinking about this moment in personal agents. It feels like a new category, a new feature, a new product, a new winner-take-all potentially. How are you viewing the market? How are you investing against it?
Well, it's a new category. It'll be a broad category. I do think it'll be very diverse and I don't think there will be a winner-take-all. Every one of the big labs or big companies, Meta included and Google included, will want to have their agents. They will have different trade-offs and I think we'll see a lot of diversity in requirements. As to Alex's comment, I think he hasn't used agents enough to understand much of that diversity that he's thinking about. Like we want to browse. The best way to browse is have the agent do the work for you and give you the options and ask you questions and be interactive with it. It's amazingly good when you can do it that way. And I think some companies will optimize for it. I think the fundamental difference will be do you want to be an advertising product or be something else. Do you want to trust Meta with your data or you want to... and some people will and they do today and I think Meta will be successful. I'm not trying to say they won't be, but there's others who will have different trade-offs. And I do think agents from startups will move faster than agents from the big companies.
Interesting. Okay. Yeah. And that makes a lot of sense. I wanted to ask you about the quote-unquote knife fight that's unfolding between agents which are right now deeply, deeply aligned with users. Right. Even fundamentally last year we went through the cycle of labs and different players launching new browsers. Which I thought was fascinating at the time because it felt like for me ChatGPT was a browser. It would replace my need to go to 20 different websites to understand like a topic or a certain product. Tab was a browser and I think that consumers like the ability to do a lot of things from one single interface without jumping around and they can ask a question about a product and then navigate to another and all these different things. But it seems like there's this something that's been basically ongoing, this fight between new agents that want to just do what the user wants in a way that's aligned to them and these platforms like Amazon that have big businesses that are under attack by these sort of new products.
I think you're right on there. There's a massive difference between a company wanting its agenda and the browser battle you referred to wasn't for the consumer. It wasn't the consumer wanted to move from Chrome to something else. It was the company saying Google controls what happens on the Chrome browser. I don't want that. So I'll have my own browser. That's a very different need than what the consumer wants. I think from the consumer's point of view, with all the assault on the consumer, the marketing assault I should say from AI and all the AI is ganging up on changing shades of blue which Meta does to make you click more or other things, TikTok plays. Do you want to be subject to the consumer or fight all the AIs with your own AI, your personal AI that only has your interests at heart? And I think that's a very, very large and important category. I think I talked about this first about four years ago, the largest opportunity being protecting the consumer against all the AI assaults and getting all their work done without it being biased. Today you can't even look at the reviews without understanding whether it was paid for by whoever's marketing it. So representing the consumer and having an AI that's loyal to the consumer, not to a particular company, a particular product, a particular point of view is going to be very, very important.
So Silicon Valley has long said consumers won't pay for things at scale. Benioff said consumers want to be entertained. They don't want to be productive. If you're building a tool that is a productivity tool, you ultimately need to sell that business as an enterprise. And yet it feels like that is going to be the most aligned model for agents because your agent can go fight different things. It can save you money. All these different things. It feels possible that this time could be different. I don't want to... maybe it's too early to say, but an agent that can get $20 a month and it saves you a couple hours a month and it... that was Mark Zuckerberg's pitch. He said Meta AI will save you money. And we're already seeing examples of people using it to cancel subscriptions that they're not using, get refunds, and so there might be some economic calculus where it looks more like paying for TurboTax than, you know, browsing Instagram Reels potentially. But how do you think about monetization?
ChatGPT with its subscription service proved the consumers will pay if there's enough value. Yeah. Massive business got built. Google always assumed consumers won't pay and that was the conventional wisdom and ChatGPT in my view destroyed that. For enough value consumers will pay. Now, saving you money is really, really valuable for most consumers. And they will pay if something does that. Knowing that it represents them, knowing every time they're about to spend some money, they can find a more efficient way to get the same thing. That's a big deal for consumers because whether you're looking in travel, whether you're looking in shopping, there's always ways to save money. Subscriptions is a great example. I have way more subscriptions than I know about. And so representing the consumer's interest and acting on their behalf as if it was an omnipotent personal assistant is what matters. And personal is the key. It's for them, the consumer, not for some company.
How do you think about moats in this era of personal agents? There's a thesis that's percolating right now that says once you install a personal agent, once you've given one personal... maybe people won't trust Meta, certain people obviously will, but the people that do and they go in that bucket and they start daily driving that particular personal agent, they give it access to their calendar, their email, their text messages, they are not going to move to another agent even if it's a little bit better. At the same time I feel like there's a risk of if you're on Instinct you can go to Meta AI and say, 'Hey, just offboard me from Instinct and set up Meta AI' or vice versa. And so it might be a little bit less of a hassle to offboard, not quite as strong of a moat as traditional social networking, but how do you see the moats question playing out in the era of personal agents?
I think offboarding will be possible. A little bit clumsy, but then if you have a good agent, you tell it to do the offboarding. And the onboarding onto the new one.
Yeah. And we saw that with OpenAI and Anthropic saying, 'Hey, go dump this prompt in the other chat app to get all your memories come and paste it over here.' And they were fighting over that because, you know, it was possible to move from one LLM to the others.
This is, I would say, the complexity of tasks, especially as companies make it hard to offboard, will no longer be a barrier. You know, if I had to look at all my information on Google or Meta and figure it out, it's complex. If I want to know... I don't want Google to know where I was on September 3rd at 5:00 p.m. They probably know and I can probably disable that, but it's hard. AI is just going to be good at doing all these complex things. And so I think that moat disappears. So what's the moat? I believe the moat will be trust in a company that develops trust with consumers and that's where Meta has a huge disadvantage. So the trust will be the moat. That'll be important. I also think completion's going to be really important. Consumers will be loyal to products that get their stuff done. Excuse the language.
Yeah. It's true. It's true. It's a good point. It's that simple. I mean, it's like we miss it a lot, but that is a good point.
And if you look at the Wajjo agent, you don't need to know what it's doing, but it knows when it's stuck. Most agents then just tell you they're stuck or don't even tell you that they're just stuck. The auto agent, for example, will pull a human in as a tool.
Oh, interesting. Use a human to complete your task. Interesting.
So the idea of trust first, your information, your most private information, what you like, what you don't, what you buy, what your preferences are. That's trust. Then there's completion. If you can get all my stuff done, I'll be loyal to you. And for the very large companies, it's very hard to have humans in the loop.
Mhm. Oh, sure. Sure. I think... Yeah. You need to scale it. I mean, it's very difficult to come over the top with, you know, a product that can service a billion users that has humans in the loop because you have to stand up that. So, there is an advantage to being small and agile and... Okay. Right.
And frankly for people who need stuff done, it's much, much better if you pay a little for the human in loop and get things done. Then, yeah. So the other thing that's really important, and you talked about the Chrome browser, not everything will be in the browser. The right agent won't be an app. It's just where do most people live? They live in their email, they live in text messaging, they live in Slack, they live in other channels. And so the agent has to operate primarily with all those channels and the app if there's one, and there may not be one. It's just in the background. So you don't have to open the app. You just text message your agent like you do with Fox from Wajjo and it takes care of it. Or sometimes you just pick up the phone and do that. And sometimes the agent just has to pick up the phone and have a conversation with the merchant. So, you know, meeting a consumer where they are. The consumer experience is really, really important. Not get stuck in we do text only, we do browser only, you have to be in our app. Users don't like those constraints.
How do you think about... I don't want to install one more app on my phone. Yeah, it's reflected in the app store. It's all just AI apps. How do you think about the way you will interface with personal agents over the next few years? If you want to send a message to me, John, will I receive that message from you? Will I think it's coming from you or will I think it's coming from your agent? Will your agent identify itself as Vinod's agent with a different name, with a different identity, a different persona perhaps, or will you want to tuck the AI beneath the fold of your digital life so that it's sending emails as you?
So, I was actually doing some ordering from Petco yesterday. And it said some things to me and I said, 'Just pretend to be me.' And it said, 'I can't do that. I won't do that.' So this goes back to trust as a key moat. Right. So I do think for both parties you have to be transparent. The agent is your agent but it's not you. And I think that's important and that'll engender trust and we'll see much better.
I want to stay on Petco. Dog, cat, bird. What type of animal are you most fond of?
There you go. What kind of dogs? What kind of dogs?
Oh, I'm a big fan of Newfoundlands. Anything.
I have a Newfoundland. I love Newfoundlands. I had two. One made it to 11. He passed away. Rest in peace. But we have another one that's going strong. About a 130 lb dog.
Lovely dog. I had two Newfoundlands that died last year. They were both around 200 lb. And so we now have a puppy, but I was trying to get a task done. I could have looked at six websites for this specialty type of dog collar after surgery. My puppy had surgery last week. It could call lots of stores and look for do they have same-day delivery, which is the key thing I was looking for. And can it pay for me and all that stuff. One of the great things is payments is going to be a real differentiator. If you can be absolutely secure and give your agent like your credit card, that's a big deal.
Mhm. And there will be lots of abuses or mistakes in that domain.
I feel comfortable with the Wajjo agent giving it my Amex card. Because it has certain types of protections that are going to be patented. So won't go into it, but I will be absolutely sure that card won't be reused for something I didn't authorize.
Yeah. Yeah. You can sort of set up payment rules if it's over this threshold. And that's actually in code. That's not something that can hallucinate. It made a mistake and the model thought it was $100. It was actually $100,000. It makes a mistake like that. So yeah, makes sense that we'll be in this blended regime of both, you know, deterministic software, stochastic software. Last thing for me, do you think we'll get to the point where users will want to give like something like power of attorney to their agents? Dan in our chat was asking about that and I've thought about this before. Just working with normal executive assistants, it's always very frustrating where you get to the point where a certain service will be like, 'Hey, they actually need you to call now. I can't sort of like take this any further.' So it feels like we'll get to that point eventually.
All these questions come back to what's the trust is the moat and will you give your power of attorney to the agent if you trust it. My bet is 5% of users will do anything. 95% will not trust giving your power of attorney to your agent today. Now in four years will that have started to happen? I suspect 40 or 50% of people will give a power of attorney to their agent by 2030, right? Because they've had enough interaction and developed enough trust. So trust becomes the key issue and this is why startups may have an advantage. Startups with great transparency will have an advantage over people who already have a brand of lack of trust.
Yeah. What's been your journey with Wajjo and Fox? Did you invest early and then this is just the right product at the right moment? Did they pitch you on this day one? Has this been years in the making? What's been the journey?
Well, we invested in it day one. In fact, we brainstormed with Shivani who's just a great technologist and a real understanding of the human factors. Those were two things that are key. So, we invested and we even brainstormed the idea. So we're really really happy with Wajjo. And she's of course deeply technical, has this other understanding of what consumers need and I call it empathy for consumers is really what you need. And say how do I best represent them? That's the right mindset. So we've been very excited about...
Do you need an access code? Is it in early access? How does actual onboarding work right now? Because it feels like if there's a human in the loop, you can't jump straight to general availability because you might get flooded, right?
You also want to grow slowly. Because you want lots of connectivity to lots of things and that builds over time as you get more users. Today, it is not free access. It's limited access by invite only. And hopefully you can give people an invite code from the company.
Yeah, we'll share out some invites to anyone in the chat, but by September 30th, it'll be a free release. So, by the end of the month, you'll see it pretty open to everybody. Awesome. Well, thank you so much for coming on, breaking it down, and yeah, great to see you. Tussling with all the different trade-offs in the modern era. Have a great rest of your day. We'll talk to you soon. And send my best to the puppy. I hope he heals up quickly. Let me tell you about Cisco, critical infrastructure for the AI era. Unlock seamless real-time experiences and new value with Cisco. There's some breaking news before we bring in our next guest. The FAA, the Federal Aviation Authority, has restricted airspace up to 10,000 ft for the Tesla Roadster reveal. Remember, we have a bet going here. Elon has teased that the next Tesla Roadster, which will be announced October 2nd, I think October 1st. He has teased that the Tesla Roadster will leave the ground. He said it might fly. There's a SpaceX collab. There's four cold gas thrusters. We had a debate. Where did you... You had six inches. You had six inches.
You were at nine and I was at six feet.
Half a foot. I thought you were at 8 in. No, I thought I was at like six feet. I want to jump it. I want it flying. I'm going 10.
I did see something. They shut down the airspace around the launch site.
That's this. They shut down. So, the FAA has issued a temporary flight restriction over SpaceX's McGregor, Texas rocket development and test facility ahead of Tesla's October 1st Roadster reveal. The restriction covers a 1.5 nautical mile radius around the site and extends from the ground up 10,000 ft in the air. That altitude is unusually high for McGregor, where restrictions tied to static engine fires and component testing have often topped out closer to 2,000 feet. So, they're going five times higher. The larger block of airspace is feeding speculation that Tesla plans to demonstrate the long-promised SpaceX package for the Roadster. Elon Musk has previously described a version of the car equipped with cold gas thrusters that could dramatically improve acceleration and potentially allow the vehicle to briefly lift off the ground. Reports earlier this year suggested Tesla was preparing a remotely operated demonstration at McGregor with the car unmanned and spectators positioned hundreds of yards away because of the expected noise, which is just a hilarious thing. You're demoing something that if you drive it within hundreds of yards of someone's house, they will be annoyed or maybe not safe. Like, how is this going to work? It's a lot of showmanship for what could wind up just being like a pretty fast sports car. But Tesla has not confirmed that the Roadster will actually leave the ground. So far, the company has only teased the event with language like 'go for launch' and imagery showing the car on what appears to be a launchpad. But the FAA restriction suggests Tesla and SpaceX are preparing for something beyond an ordinary static presentation. Whether that means the Roadster will actually hover on October 1st remains unknown, but the FAA has now reserved a substantial column of Texas airspace for whatever happens. I wanted like an F-35. I wanted to do full VTOL. If I want to park, just bring me up, plop me down. I don't know. I'm excited. I think he's got to deliver something. It's been nine years since he took the deposit. Are you going to put a deposit down?
I think it's a deposit down $45,000 due within like 10 days. And it feels like it's kind of high optionality because if it's incredible that so many people are going to flood in after the fact, but unlike waiting nine years, you can just wait until October 1st. If it's a flop, you can just get a refund. So, for locking up $50,000 for 10 days, yeah, I think you should do it.
It's pretty tempting. It's pretty tempting. I need to really read the fine print or have AI read the fine print to understand if there's a chance I get hosed, but there's a chance that you get one of the first ones. Not even, oh, it pops financially and it resells. Just the fact that it could be a really special and cool car that's really fun to have. And if they bring it in with the final price of 250, there's a lot that they can do to impress at that price range. If it comes in at 800, they're competing with a lot of... If it's just a sports car that looks like a Cybertruck, it'll be a huge hit. If it's 250, if it's 800, now you're competing with new and new has a V10, V8, something like that. Like it needs to be Cybertruck, Huracan, Cybertruck but also 250 because then you're like wow this is crazy this is competitive and that's what Tesla has been for a long time. The Model S was really, really performant, delivered a lot of value for a pretty decent price. Same thing with the Model Y, it's still like a value play even though it is a relatively mid-market expensive car. Well, I bet you Aaron has his pre-order down from like nine.
We're going to find out. That's going to be the first question. We're bringing in Aaron Levie from Box. He's the co-founder and CEO. Welcome back to the show. Did you deposit for a Roadster 9 years ago deposit?
I didn't do 9 years ago. I was very tempted this time around though. I did go down the rabbit hole. There's these great YouTube, you know, kind of speculative fiction. Yeah. Sleuth folks that... Did you see the thing if you... if you white out the video it says where we are going. Did you see this?
Is that a reference to Back to the Future?
So it might go 88 miles an hour. That would be huge.
Either that or so. Yeah. So we're... but there's a whole... I went and watched a 20 minute video. I'm very excited. I'm glad that you...
What's your theory, though? What's your theory?
I don't know. I mean obviously Elon would make a flying car so it's... I think we're all waiting for that. Now I guess the question is like practically what is it, what does that do if you literally can't fly it? But...
You can fly it if you're within 1.5 nautical miles of Starbase Texas and the FAA clears you I guess. But I wonder if there's some loophole where like if you're on your own private property, can you get off the ground and you could have, you know, those people have those VTOL things. Yeah. No, no, no. I was going to say like, you know, you drive onto it and then it spins your car around. Yeah. Yeah. Turn. Could he make it? Yeah. Turntable. So we could make it like, you know, if your garage is kind of tight, you want to turn around little indoor. We may have reached the point though where Elon is building things for his personal life that exceed our normal zoning laws of mere mortals.
There's going to be a lot of people with deposits that are like, 'You're telling me I need 100 acres of private land now? I was told that this would go 0 to 60 in 2 seconds.' And the craziest thing is the original specs were 0 to 60 in 2 seconds. Cars already do that. The Model S Plaid did that. Not crazy. 250K. Okay, that's like a reasonable price point for a supercar these days. But over 600 miles of range, which has actually not been done by any EV yet. And in a smaller vehicle, it's not like it's going to be huge. So, very, very uncertain however. I mean, they seem to be doing things with batteries that I never thought would be possible with batteries. Like I was always told like 20 years ago that, you know, we'll never get more than 300 miles in a car. So you just wake up one day and I guess it's 600. So yeah, I mean I guess if you launch with the cold gas thrusters, put it in neutral, coast for 100 miles of range.
Maybe that's exactly... Yeah, they get you the first 300 miles with... Yeah. 600 downhill as long as you just cruise them. Anyway, how are things going at Box? Roller coaster ride? SaaS apocalypse fully cancelled. We were very happy to see that. A bunch of... We were early to cancel. We bought them tickets.
You guys cancelled it within like 3 hours.
Yeah. No, we were like, 'This is not happening.' No, but we did have a newsletter that went out like during the depths and I believe we were the first people. SaaS apocalypse is cancelled. Yeah, because... I actually, you know, you almost couldn't cancel it too soon though because you had to have the full trough.
Yeah. And you had to have like full capitulation on, okay, it turns out that there is actually software that is powering the world and these systems of record are incredibly robust and actually the more agents that you have running on these systems, in fact, the more important the backbone becomes. Everybody always sort of made this claim that once you have all these agents then actually all they really need is a database underneath. But the problem is that database has to be coordinated against possibly thousands of end users and then possibly tens of thousands of agents. And so the access controls, the business logic, being able to actually render the experience, the security obviously, you see OpenAI, Hugging Face, like these core systems are actually going to become more important in a world of all of these agentic workloads. So it's good to see the market, I think, paying more attention to that reality. And it's obviously driven by, I think, the continued performance of these software products doing well in the market.
Yeah. I mean, even over the weekend, I was generating AI videos. I was trying to get them from my PC to my phone, and at no point was I like, build a new storage system for these videos, even if it could do it. I'm like, I don't want to waste a prompt. I don't want to waste a second doing that. Let's use what I already use. Let's use all the tools that are already built out that are robust and I can share links and all these things. I don't need you to one-shot some shoddy clone. It's a waste of time. It's a waste of money. It's a waste of energy. Bunch of things. But in terms of resetting the narrative with the market, the street, investors, employees, everyone. How much of it is going back to, okay, you thought this thing was existential. You thought we were going to be in so much trouble. We're not. It's business as usual. Versus there might be still some threats but there's more opportunity and that's actually outweighing the threats and so that's why the company's performing well in the markets over the last couple months, for example.
Yeah, very much not business as usual I think for anybody in software. So, like this is like... I don't know of a single company still being run the way that it was run maybe, you know, 18 months ago. I mean everybody is running insanely fast into the future and then I think there still will clearly be some parts of the market that don't capture as much of the value as they did previously and then there are going to be some parts of the market that capture actually a ton more value than they did previously because of what agents can do. I mean you look at, you know, infrastructure providers as an example, like the Cloudflares of the world, totally on fire because agents need sandboxes, they need compute, they need network, they need gateways, like great business. If you look at...
DataBricks or Snowflake, agents need to be able to triage against large sets of data that you need to be able to pull together in one kind of common system. So doing fantastically well for us, we have reaccelerated growth far past our internal plans because it turns out that enterprises need core systems to be able to manage their unstructured data, and agents need to be able to work with that unstructured data to make decisions or move information through a workflow, whether it's all of your contracts or your research materials or marketing assets or financial documents. And again, like it might be easy in a personal productivity sense to be like, okay, well, I can just have that data in some cloud system or even in my local device. Now imagine an enterprise with 10,000 employees. How do you ensure that those agents continue to go after the actual canonical records, the actual documents that are the real sources of truth and that are the authoritative versions of that data? That means you need platforms. And we're obviously competing for how do you become the platform for all of this unstructured data. But I think there's a lot of value being created really across these ecosystems, and I would say right now my view would be it has unequivocally been a net positive for software, even with some of the spaces being under pressure, simply because of how much more work agents are doing on these systems and the need to be able to provide the guardrails and be able to be the traffic cop for what those agents are doing, and to be able to deploy these agents natively within your own agent harness, in your own system of record, and all of that is creating more value for these platforms.
At a system level, what is the biggest lever that you want to pull? Is it like speed of interaction with agents? Because in the previous paradigm, search and file retrieval, if it took a couple minutes, somebody's drinking coffee, they're checking their email, maybe it's not as crucial. But if my agent is hanging because I'm churning through a bunch of records or files, that could actually be more annoying. Or is it just the volume of content? Because even an employee, if they're not totally slopping it up, they're probably producing more files, more content, more video, everything's just multiplying. So where have you been like, okay, this is the lever we need to really pull to make sure that agents love the Box experience?
Yeah, I mean, I think realistically all of the above. First of all, there's an explosion in just content creation in general. Agents are producing content obviously at an unbelievable rate. And you know, if you look at...
Have you gone through this? I want to talk about that. Let's finish what you're saying.
Sure. Sorry. Yeah. There's just data more than ever. And you're going to need to govern it. You're going to need to store it. You need to manage it just as if a person created it. And so all of that is obviously like the core backbone system. We're doing a lot of work at the second layer, which is we effectively think about it as sort of context retrieval for your unstructured data. But obviously, if you did a regular search two or three years ago and you compared it to a search today, when we actually can capture embeddings of those documents, our agents do re-rankings of the search results. We search in parallel, so we can run multiple versions of your query at once. Your chance of finding your document not only faster but just generally more accurately is way higher today than it was a couple years ago because we not only have all these other signals but we can parallelize the process much better. And so an agent from the outside coming in and saying, hey, I'm looking for this record, you're able to now much more accurately achieve that result, which is fantastic. And then I think the thing that we're also spending a lot more time on is how do you better secure data in this world of infinitely more agents than what we had three or five years ago. The OpenHuggingFace incident has many interesting ramifications to how we build agents, how we ensure the safety and alignment of them, but has very real-world implications on the security of our system. So while this was more of a production system from HuggingFace that was under threat, you could imagine the internal environment of an enterprise. If you have all of your critical contracts or financial documents in an environment that agents can eventually get their way into, enterprises are going to need to care deeply about how do you protect that information. So we're imagining and reimagining all the ways that you would protect data in a world where you had agent swarms being deployed against your data environment, and how do you go and defend and protect that type of risk. So lots of layers to this problem, but really just through the lens of imagine you have 100 times more agents than people, how would you work with, store, access, search, and secure your data in that kind of world? That's the platform we're building.
Do you think we need to move away from the word sandbox? Because I don't know if you have kids, but if you've ever taken a five-year-old and plopped them down in a sandbox, at no point should you as a parent believe that that kid will stay in that sandbox. The sand is going to go everywhere. The kid's going to be on the playground two seconds later on the swing set, they're going to be running around crazy. Sandboxes have never been secure environments. They're not jails.
Yeah, I think I don't know who that terminology is. It's running its course.
I think it's running its course on personal agents. How are you processing the new race to be the most helpful assistant ever? And then specifically, I want to understand how you think all the agents versus walled gardens war is going to evolve, just because it seems incredibly obvious that Amazon will fight to the death to maintain their like $75 billion ad business and the health of it. But what's your read?
Yeah. Super, super fun how fast-moving the space is. I want to first apologize because I honestly did not have the imagination on the consumer front. And so I was pretty skeptical of what the breakthrough would look like on the consumer end. And it's funny because it was obviously staring at us with OpenClaw. But just the form factor and the packaging and getting it all kind of in the right way that both Muse and Instinct have done, and obviously there's going to be more entries into the space, has just been phenomenal. And then the capability progress on computer use and browser use is just amazing. And I do think I have a wish list of what we need architecturally as an industry. Like you need to eventually be able to have some better password sharing dynamic. If I can in one click be able to get into any of these products in a very secure way and there's a trusted handshake between the agentic system and the tools that I'm using, we're going to need that. Payments are largely solved because of obviously the work that Stripe and others have done. So I think that's pretty seamless, but really it's the authentication that we still need to work on as you're kind of ramping up the use of these agents. But I think it's going to be an incredible market. The ability to now hand off the whole problem to an agent as opposed to just using it merely as a research assistant, I think is going to be a breakthrough for consumer use cases. I'm very excited about what this can mean for all new business models in the economy. I think there's a, you know, the Catrini post, you know, five months ago, four months ago or whatever, had some of the more, you know, kind of risky to capitalism elements in the market, but you can easily look at the exact other end, which is I might now be a consumer of five times the amount of local service providers as a result of these types of systems. Like if I could lower the barrier to just like, I need flowers like right now, and it handles the whole thing, I would, I mean, at least in my personal experience, I'm going to grow my usage of the local economy as a result of this. And so very interesting to see, you know, you might have, I think some...
Cutting out the middleman and installing an even more powerful middleman.
Yeah. Or layers and layers of middlemen. It might be middlemen all the way down. I mean, I like Yelp as an example is like, how do you sort of triage the local economy as a graph?
Well, there's something even with how good the internet has gotten in so many ways, like there's still specifically local providers where I think using flowers as an example, like how many, the place that I order flowers from, it's a good reminder to everyone to order flowers for your significant other.
Five months until Valentine's Day.
Valentine's Day starts now. You're never too soon for flowers. No, but it's like the service is, they make amazing bouquets, but the process of actually making an order is like so terrible. It's so painful. So painful.
Yeah, just that little bit of friction. And so people say, okay, consumers don't want productivity. They don't care about productivity is what a lot of the bears are saying right now. But they also hate friction.
And so what's going to win out? You know, I think...
I think that's the thing is I'm so utilitarian and productivity-pilled that I shared the same view of like consumers don't sort of value their time in the same way, but I think it's just then not on that dimension. It's like the pitch is not to save you time. It's just to do things that you've probably added to your remit. And then there are lots of ways which are like that, like on one hand they're time-saving, but on the other hand they are just like literally doing more. As a consumer, like, you know, we have three kids. We're always signing up for more things. There's more camps. There's more, you know...
I don't know, what was that one for three kids? It's just very good. Congratulations.
We can even take it a step up. We have nine kids.
We have nine kids between us.
We might have nine kids in the future.
So, you know, you have camps and museums and this and this thing and you're renting that thing. And so it's not even like you're saving time. Like if you could just have an agent do a lot of these activities for you.
Yeah, stress relief. And then you actually just do more with your kids and it's like, it's fantastic. We totally... we went to, there's these Japanese tea gardens in San Francisco and I had an agent...
Oh my God, what is that one? That's for Japanese tea garden.
Excited to castle in the castle.
So, and we had an agent book the whole thing and it was like, and we just like added that to our plate that day because it was that much easier to go and execute. So I like the idea of more of the kind of consumer abundance use cases. And so I'm very excited by this as just even more volume in the economy of just buying and selling. I mean, you can imagine the Thumbtack, Yelp, TaskRabbit use cases of, hey, I need help right now to go install this thing. And it's just like all new forms of activity happening. So I'm a big fan of the personal agents. I think it's just full net growth in the consumer economy.
