Deel–Rippling Update, Microsoft Bets on Agents | Shawn “swyx” Wang, Augustus Doricko, Brad Porter, Kian Sadeghi, Kathleen McMahon, Chungin “Roy” Lee

29:56Follow the app TVPN on X. I don't know why you would see this and not be subscribed. There are plenty of people. But there might be people in the RSS feed or something. Maybe you've been, maybe it's like, oh, it's too much content. A, let the algorithm sort it out. But B, just give us a follow just for a little bit. We're so close to 64 ,000, which is another doubling. We'll be doing something to celebrate. So thank you. and in other news Snowflake is buying Crunchy data for$250 million this is from the Wall Street Journal the cloud data company aims to attract customers who want to build their own artificial intelligence agents let's give it up for a base hit for these investors yes it probably didn't return the investors fund entirely but it's still a fantastic outcome hopefully for the team and so we got it we have to do like a database day or start talking to some of these database folks because Snowflake, Databricks, Palantir, they're all making serious moves in this space and they're all kind of moving to different layers of the stack.

31:01Databricks is now more of this data lake, like data unification layer and then Palantir sits on top as like the ontology layer, actually understanding how all of the different data interacts. And so they're kind of playing nice and Snowflake and Palantir used to be kind of comped directly to each other, but they've diverged in the public markets. But still, Snowflake's like a fantastic and incredible story of an incubation at Center Hill Ventures that went massive and IPO'd and went into the tens of billions of dollars. Databricks is also up there in the tens of billions of dollars. A lot of people are waiting for them to IPO.

31:41And so it's a very interesting dynamic. Now, Databricks bought Neon, a similar database startup, in a deal valued for about a billion. Now, Snowflake is buying Crunchy Data. Great name. Yeah, Crunchy Data. The naming schemes in the enterprise SaaS, like deep down, are just wild and entertaining. Like Datadog remains one of my favorite startup names of all time. I mean, yeah. It's so funny. data dog i mean you have uh last yc demo day we met the founder building pig pig was great too we gotta check in on pig we gotta check in on pig i don't exactly remember what they do but yeah but i but i remember the name yeah yeah it'll only be a matter of time until we see like the i'm now searching pig ai luckily i was able to find it pig.dev there we go an api to launch and automate windows desktops let's hear it for pig let's hear it for pig just a great name i think you know if pig is successful it will inspire a generation of of companies named after animals yeah monkey goose yeah horse horse.ai we don't actually we don't have any animal name companies although we do want uh we do want animals to start sleeping on eight sleeps yeah that's something we're pushing for.

33:02We think animal testing is going to make a big comeback in the mattress market. Eight sleeps are so comfortable. We're going to have, your dog should be sleeping on eight sleep. How'd you sleep last night? I think I had a rough night. I was up at five on the grind. Let's see. I actually didn't, I didn't get the hours in. You didn't get the hours in? I slept for six hours and 50 minutes. I got an 80. I got an 84. Oh, you spoke to me again. It's so close. I got six and a half. but so close but so far yeah it happens it happens to the best of us anyway um it's pretty funny my um uh my son uh like came in our room at probably like 2 a.m i was i was half asleep but then he just slept like um uh uh like perpendicular perpendicular to us yeah so he was getting two different temperatures like half his body was getting you know my wife's temperature half the other um which had to been a funny i mean just weird weird experience but he's he was sleeping well he was sound asleep when i when i left yeah yeah i remember showing showing my son the uh the eight sleep app and i just had you know you can see like the left side and the right side and like what the temperatures are yeah he was like and me for the center and we're like no they don't make that they don't make that yet also go back in your go back in your um i had to pull up startups named after animals okay because there are actually a few male chimp post gator mule soft post hog post hog post hog is like pig it's the same thing a hog and a pig post hog is one of the great things ever post yeah i mean it's it's an iconic it's like post modern like post hog like we are post hog we are now in domesticated hogs which are pigs so pig in many way is the post post hog yeah i don't even know if they do the same thing uh task rabbit survey monkey hippo insurance it was a big it was a zebra yeah attractive okay okay now we're going now we're hallucinating we're hallucinating fat llama a peer-to-peer rental platform lunch badger a cloud services badger these are hallucinated this is not real no it's real it's in a crunch base.

35:15Let's go. I mean, if it's in crunch base, it's, well, is it in crunchy data? The thing that stands out about pig is that it doesn't add anything before or after. No, no, no, no. It's like ramp. Yeah. Or like Vanta automate compliance, manage risk, prove trust continuously. Vanta's trust management platform takes the manual work out of your security and compliance process and replaces it with continuous automation. Whether you're pursuing your first framework or managing a complex program, uh, back to crunchy data. Crunchy data has roughly a hundred employees. scale pretty quickly. They'll join the company once the deal closes, which is expected to close in the next couple of weeks.

35:47Crunchy data will be part of an offering called Snowflake Postgres. The vision here is that Snowflake Postgres will simplify how developers build, deploy, and scale agents and apps. Very relevant to that Microsoft story we were talking about earlier. With that in mind, it was important to acquire a company that was not just engineering and quick experimentation. Crunchy data is a cloud database cloud-based beta database provider that helps large businesses and government agencies use postgres postgres without needing to manage infrastructure themselves i've used postgres before there are some it's crazy i didn't realize the french data was founded in 2012 wow overnight success overnight success you love to see it congrats on the 250 million dollar sale how much did they raise i wonder um i'll find out there's a couple other um deals snowflakes offers a platform for storing, organizing, and analyzing data across multiple cloud providers, including AWS, Azure, and Google Cloud.

36:43The company which went public in 2020 grew rapidly during the pandemic as more companies migrated their data to cloud storage from on-premise data warehouses, which is, it's crazy that we're still in like the cloud migration era from on-prem data warehouses. The cloud data has existed for a decade and still happening. And if the AI rollout is any any similar or if it tracks similarly. It's like we could be talking about agents for the next decade and just continually rolling these products out in a very, very slow takeoff scenario. Crunchy data only raised$14 million. Wow, there we go. Very efficient.

37:21Very efficient. Well, speaking of other efficient businesses, there's a story from our friend Chris Best in the information. He says Apple's App Store changes, quote, have been fantastic. stick. And this is why I wanted to highlight the efficiency. He said Substack was accidentally cashflow positive in the first quarter of this year. Fantastic. Growth has been translated to revenue. Best said the company, which was founded in 2017, was accidentally cashflow positive during the first quarter of 2025. That said, Substack is not focused on profitability right now, he said, as it hopes to continue investing in growth.

38:02This phone. Said that Substack apps' users are heavily in the U.S., where Apple has been ordered by federal court in Northern California to allow app owners to offer users alternative payment mechanisms, allowing apps to bypass Apple's 15 % to 30 % fee for in-app purchases. So, had Apple been taking... Yes. If you went in the app and then you subscribe to a sub stack and you use the in-app payments flow, they would take 30 % of that on an ongoing basis. I didn't even know that was possible. I've never signed up for a sub stack in the app. Yep. Yep. And so that's crazy because this, this, yeah, I mean, it's a digital product, right?

38:46And so it is. Well, it's interesting because Audible has found a way to get around this by. Credits and stuff. Credit system, which is just so annoying. Yeah, I think that that's a bigger negotiation because it's Amazon. Yeah And I think I think so I think Apple had more leverage over sub stack But we should have Chris back on the show and ask about how it actually works But yeah, I mean if you're I feel like most sub stack users would be actually fine going through a web flow Yeah, it's not as much of like an impulse purchase you really you know You're you're you have a relationship with the sub stacker that you're subscribing to you kind of understand and there's it's pretty easy to message I feel like the average substack user probably understands a little bit more about the Apple App Store tax desktop respecters exactly a lot of corporate athletes Corporate you have people that are sitting in front of their computer all day long Yep, and what better place to buy things than the computer buying things on the computer is really Historically and still today just a fantastic experience.

39:46It's fantastic Especially when you have your sales tax automated with Numeral that's right Numeral HQ put your sales tax on autopilot spend less than five minutes per month on sales tax compliance. You may be able to avoid the Apple tax, but you will not be able to avoid state sales tax on software and consumer goods. Sales tax CGI. Yes. Anyway, Substack now has more than 50 creators who are making millions of dollars per year on the platform. Best said, the platform overall has more than 5 million paid subscribers. So yeah, it's$120 million. I didn't realize they raised so much, but that's great that they're profitable.

40:27Yes. I mean, I, I think this is such a great story where they, this is a business that, that raised ahead of where they were. Sure. And then a lot of people wrote them off because they were like, wait, this is just a media business with a take, you know, a take rate and, and all this stuff. And, and what Chris has been able to do over the past year, you know when we had him on the show i was i was kind of asking you know has it felt like substack has broken through really into culture yep uh becoming like a real brand itself outside of you know twitter right because they basically got like elon came for them hard totally and that was like somewhat warranted because they launched like a competitive product i thought the competitive product happened like as a response to the link band i thought that was the sequence of events it It was Lynx getting deprioritized somewhat and then Substack launching a post-competitor, which I don't even know if I agree that that was the right move.

41:30I haven't really played with that product. It seems obvious that X was going to ban Lynx either way. Yeah. That hurt Substack. Yeah. It hurt the writers on Substack. Sure. And was the right decision for X. Yeah. um but uh but yeah so there was a link suppression um and then and then uh and then sub stack hamish uh one of the co-founders called elon musk a propagandist with more conflicts of interest than el chapo spicy um but uh ultimately um interesting i i had never i never I never picked this up. Apparently, Musk had made some type of proposal to buy Substack in 2023. Yep, but probably not clearing the Substack.

42:21After that, they restricted links. And Twitter formerly had bought another publishing platform. I don't remember the name, but they rolled that in. And then I think they closed it down. And now you just have the ability to write pretty long posts. You can post articles, X articles, directly on X. And I think that it's weird because we have yet to see, like we are an example of a media business or a show that just said we will play the game by Elon's rules on X. So we don't think about links at all. We think about what are the products that X loves? text posts, memes, images, video uploads, live streams.

43:11And we do those very, very well. And that's what we focus on. And we don't, so we're not intention. We're like leaning into what the platform loves. We should actually consider doing articles on X. The question is, what would it look like if you tried to build a sub stack like business? Like you're just a writer and your output is articles and you use X's subscription tools and the articles feed to have that experience of one free article a week and then one paid a week and you try and build up the book of subscription business on your X account. I've seen some people that have subscriptions turned on.

43:49I think I actually technically have subscriptions turned on. I think only Gary Tan is like the one who subscribes to me. Shout out Gary Tan. Thank you. Let's give it up for Gary. He just throws me a couple of clicks. Yeah, ultimately the thing that's interesting about Substack is they were effectively benefited from Zerp in terms of accessing capital. And people started to write them off because I think at times their growth was not best in class. They were just sort of chugging along. And if he had originally had this idea of, hey, we're going to build this publishing platform that's going to allow independent sort of citizen journalism and writing to flourish, and then we're going to launch this sort of social network with streaming and all this stuff, that at certain times would have been hard to believe, but they actually have executed that to a T.

44:37And now when you go on Substack, it does feel like a social network, you know, based on, you know, email as a sort of backbone. So I'm excited to follow their progress more and eventually get really set up on Substack ourselves. Yeah. So next up, we have Augustus Dorico coming in. There's an article in Semaphore today. China boosts use of cloud seeding to combat droughts, and we wanted to have him come in and break it down for us. The headline is, China stepped up cloud seeding in the face of severe drought. The country's grain-growing regions in the north have been parched for months, leading to concerns over the harvest.

45:17Though some scientists are skeptical over cloud seeding's effectiveness and environmental impact, several countries have begun deploying it. China is already the world's leading user of weather modification, firing chemical compounds into clouds to spark precipitation and has conducted more, conducted 20 percent more than by this time last year, apparently causing a one third increase in rainfall. That seems pretty significant. So let's bring in Augustus DeRico and have him break it down for us. How are you doing, Augustus? Good to hear from you. I don't know. That sounds like a cruise ship or something.

45:52Yeah. Yeah. Welcome. of that's that's the sound that that august that's in augustus wake you know yeah it's great to have you yeah thanks man that and size gongs that and size gongs um update soon wait you gotta listen to this one journalists on the horizon stand by great who was the guy that was responsible for posting the uh state try to stay focused on the mission gif a while back when i got that text i got like a hard eyes text about like oh i like your appearance on tbpn social media intern there Yeah, yeah, yeah. Yeah, the team's growing over here. We got a good crew coming together. There's vibe coding happening over in that part of the studio.

46:31There's a lot of production stuff going on. It's been a fun time. Sweet. So I'm here to ring the alarm bell. Totally transparently, there are really big problems on the horizon, both with China's domestic weather modification program and then their international one. the article that you read or that you're referencing talks about how China is actively retrofitting their wing long twos right talk about like nominative determinism cool name for a drone but it is the essentially Chinese equivalent of the MQ-9 Reaper and they're using it for cloud seeding and weather modification operations all across the country a lot of that is to fill up the snowpack in Tibet and then use that as a natural water tower for runoff for all of their urban, industrial, agricultural, and environmental assets in China.

47:21So that's that unto itself is insane to recontextualize people. The Chinese meteorological administration has about a$300 million budget for weather modification. They have 38 ,000 employees exclusively working on weather modification. And they have two universities that offer bachelor's degrees in weather engineering, not meteorology, not atmospheric science, specifically engineering the weather. So they are driving extraordinarily hard on this domestically, just for their own water supply, just to green deserts, just to keep their cities and industry humming. The problem is the international implications of this, right?

47:54We are in a, and like I critique people all the time for saber rattling with China needlessly, but the Wing Wong 2 has been sold and is being operated in countries across the world, namely Saudi Arabia and Egypt for defense applications, right? So that has its problems for defense, but also for weather modification. The CMA has explicitly said they want to export cloud seeding as a means of soft power to control water supply and weather across the world. They already collaborate very closely with the Thai Royal Rainmaking Department, and they can easily retrofit these drones that they've sold for defense applications.

48:31The Royal Rainmaking Department. Wow. Great name. Great name. But they can easily retrofit these drones for weather mod as well across the world. And then not only control the shipping and receiving and the ports, not just the energy infrastructure. Nice. But also the water supply and weather. And President Lyndon B. Johnson said, whoever controls the weather controls the world. And right now we're trending towards a world where China controls the weather and subsequently the world. And it's really Rainmaker Technology Corporation representing the United States against the CMA. So President Donald Trump, if you can hear us, Secretary Rubio, if you can hear us, the State Department should be involved in this soft power conflict on weather modification internationally.

49:15How are you thinking about the current pitch for weather modification? Because it feels like this news out of China has a few different hooks. As well as agriculture and drinking water, much of China is reliant on hydropower for electricity. Sichuan in the Southwest gets 80 % of its power from dams, meaning that droughts can lead to electricity shortages. I mean, I know we have the Hoover Dam, but is hydroelectric power important in America? Obviously, power is top of mind for everyone. But in America, it feels like the narrative has shifted to nuclear and solar. But is there a world where we could be getting more out of our existing hydroelectric assets and there's maybe a narrative there or not?

49:58A hundred percent, right? Nuclear is awesome. And I'm super excited for Ballaratomics to turn a thousand reactors online in their gigasites and produce the world's energy. But it's going to take at least a year for that to start. And it'll probably take years longer still until that's our main form of stable baseload. Solar is great, but we have nighttime still and Reflect Orbital hasn't solved for that problem. So we need stable baseload and we need clean baseload. And hydroelectric power across hundreds of dams in the American West produces hundreds of gigawatt hours and we can refill our own dams to increase more stable clean base load with cloud seeding where these dams are drying up right now we can increase supply you know 80 of uh you mentioned sichuan 80 of columbia's power comes from hydroelectric as well and they're going through a drought right now and so they have rolling blackouts because there's not enough power there both domestically for our own energy production we could use cloud seeding to produce more hydroelectric.

50:54And then internationally, again, as a means of collaborating with other countries, let's call it, and ensuring that they have American interests in mind, we can produce more water and hydro for them. Can you talk about the Chinese approach to cloud seeding versus what you're doing in the American approach? I feel like a lot of times when we see a competitive dynamic emerge between China and America, there's only a small tweak between the way Instagram reels are served versus TikTok or DJI drones versus GoPros. It's usually just the scale and efficiency and reliability of Chinese technology, but there isn't usually that much of a shift in the underlying strategy.

51:35Are they using the same chemicals? Do you think we should be using different chemicals from their mix? Are they using different drones or are they shelling this stuff into the atmosphere with howitzers? Is there anything that we can learn that might not be IP protected that we could safely port back? Is there anything that we should change based on what we're hearing from over there? China is throwing the kitchen sink at weather modification research. So they're doing drone-based aerial cloud seeding. They're doing ground generator-based cloud seeding. They're doing acoustic cloud seeding research, meaning they have these huge 130-decibel speaker systems where they just blast it at clouds.

52:15They put them all in Tibet. So even though it's like destroying the ears of Tibetan villagers, they're trying to shake water out of the clouds. So that they have a bunch of other photonic stuff. They're doing a ton of research, but really the big and important aspect of this is their sophisticated military retrofit of drones for long endurance missions, their radar research for detecting phase change in cloud. And then lastly, I think the thing that they have like the most serious edge on the United States than anybody else in the world in is their ice nucleation agent and their particulate. They're doing a bunch of nanoparticle design.

52:55So super, super small scale particle coating, it's titanium dioxide on top of these salt crystals, among other things that are way more efficient at nucleating ice and subsequently creating snow or condensing stuff in cloud. Rainmaker's doing research into that right now, but that's where China far and away has the biggest lead on the nanoparticles that they're using. How about your challenges at home with various states and regulators? What's the update there?

53:25So 31 states proposed legislation to ban all forms of weather modification this year. Almost all of them dropped that legislation because one.

53:38because of you because uh was it did you we did me and the rainmaker team did a lot of work around state capitals i've got like a regular barber shop in tallahassee and a few other state capitals um to tune the mullet mullet up before i testify um but uh the one state that did ban it was florida florida made weather modification a class two felony um so if i were to work there I'd go to prison for five years. And that, I think, unto itself is not the huge problem. Fine. Sure, you're depriving Floridian cattle ranchers and orchards from having as much water as they want. And they do have wildfires and droughts.

54:15So it hurts the state of Florida. But the real problem is the canary in the coal mine in terms of American political sentiment, particularly Republican political sentiment. Right. There's this whole conversation around the tech right and who is pro-innovation and who is not. Is the Trump administration pro-innovation? seems so for the most part. But unless we have clear federal regulation on weather modification and a public stance in favor of this, then we're going to lose control of the weather to China. Yeah. What about desalination? It's something that we were doing in America. We kind of fell off.

54:50It feels like the nuclear story. And it just feels like I'm going to hear a story in the next few years of like, oh, yeah, China just figured it out. And now they have like a bunch of desalination plants and we're behind on that too. Are you tracking it at all? So desalination is largely held up by the California Coastal Commission. And then like HOAs basically that block the construction of desal. An interesting stat that - Because it smells bad? No, just because it looks ugly. Desal smells fine. It's just like a bit, well, it looks, I should say - It looks beautiful oil refinery, which I happen to be a fan of personally, but folks in Newport Beach less so.

55:35Desal is great. And if we come up with some really sophisticated new reverse osmosis membranes or catalytic desalination methods with good electrical engineering, we can make it more efficient. But the problem with desal still is that we have to move that water from the coasts. Like it's a non-starter for Nevada or Colorado or Utah to get desalinated water. What you can do though is cloud seed, obviously, right? You can produce water anywhere where there are clouds with our tech. Yeah. I remember Blake Masters was saying when he was running in Arizona, he was saying like the future of California is nuclear power desal.

56:12And then we need to reroute the Colorado River to hydrate the inner states. Is there a world where you could build? I mean, you mentioned it looks like an oil and gas refinery. Can you build like an offshore oil rig essentially? Or is it just like it's not economically dense enough? Water is not the same as oil, so you're not going to be able to pay to put it on a truck and then bring it in. It just doesn't make sense to do it that way. Yeah, exactly. It must flow. I was talking to some commodities traders the other day, and I was trying to like come up with some crazy derivatives for water. Water.

56:47It's not right there yet. You know, it's it's it's it's sense for a barrel. Right. And it's just it's almost as heavy as oil. So you can't convey this. You know, 13 percent of all of the electricity in California is used just moving water around. What? That's crazy. Wow. Yeah. Like the Central Valley exists just because we pump all of that stuff from the Sacramento Delta and American River down into the Valley and elsewhere. Like there's this huge, huge unknown problem, which is like, because we don't have enough water, we have to dedicate so much of our energy resources just to moving it around where we could just be producing in the Sierras.

57:25And that's what Rainmaker's doing. Interesting. Anything else you're tracking from China? Um, so I guess one, one thing that I will say in terms of like this soft conflict is we had a customer meeting, um, in the middle East, um, and a day before we had the scheduled meeting, they said, Hey, sorry, we have to go to a last minute trip to China, um, to go talk about the, uh, Wing Long 2. Um, so, uh, we're trying to qualify Rainmakers vehicles in the middle East right now. Um, so that we are clearly at parity with their system capabilities and then can be selected over the Chinese. Yeah, the challenge, if you look at the precedent for sort of state-backed companies out of China, they're willing to sell at a loss for years.

58:12And so I have no insight into whether they would try to attempt that in cloud seeding, but it wouldn't exactly be surprising, which I'm sure is a competitive dynamic that you're thinking about. Every single international office that we go into to talk about our work, we see a stack this high of purchase orders from China and then one that's like two pages thin that's from American companies. So, yeah, it's absolutely a problem. Wow. Part of this article highlights this idea that grain growing regions in China have been parched this month or four months. Are we in a particular global drought? Is this unique to China?

58:57Are we experiencing drought in America? What is the state of drought generally? you mentioned that there are different pockets around the globe, but is this a particularly bad year? Are the trend lines bad overall? I mean, we saw the fires and it felt very visceral in California and Los Angeles, but you never really know when you zoom out where we are on the trend line. So we had out of distribution high amounts of precipitation this past year in California. Yeah. And even still, we had the wildfires, right? You know, the thing that I think is a good reference point is the California Department of Natural Resources water supply strategy.

59:39They explicitly plan for half a million to a million acres of farmland in the Central Valley to turn into desert in the next five years, just because there's not enough water. So even when we have boom years and the reservoirs are all full and there's tons of snowpack and everybody gets to go skiing in Tahoe, there's not enough water for current demand. And that's in part just because population is growing, right? Like we built the Central Valley Water Project to turn the Central Valley from a desert and swamp into the most productive agricultural region in the world. And we wouldn't have the US population that we do now if it weren't for that water project.

1:00:11And so we just need to produce more if we want to maintain agricultural and economic growth. Makes sense. Well, good luck out there. We hope you can strike some big deals in the back of this news uh you know we got to be competitive thank you for fighting on america's behalf we'll talk to you soon come back soon later cheers see you bye next up we have brad from cobot coming in the studio talking about collaborative robots i'm very interested to talk about the the sim to real gap which we covered yesterday uh we will welcome brad to the studio how are you doing welcome to the show. Thank you for joining.

1:00:47What's new? Oh, fantastic outfit. There we go. Welcome to the show. Amazing. Looking great. What's the occasion? First, introduce yourself, please, but explain why the fantastic outfit today. Are you doing real work? Yeah. That's right. Yeah. Hi, I'm Brad Porter. I'm founder and CEO of Collabor Robotics. Yeah, this is how you tell whether Whether a robotic company is really in production or in the field, if they're wearing their safety vests. And I was at our deployment with Maresk yesterday. And so I had this handy and thought I'd bring it out for you guys. Yeah, so I mean, to the degree that you can talk about it, what exactly are you doing for Maresk?

1:01:33That sounds like important work. Yeah, we're helping them in transload operations, in unloading ocean freight containers and loading out onto tractor trailers. They load industrial carts full of up to 1 ,500 pounds worth of generally boxes of product for retailers. And then we help with moving the carts, because moving those heavy carts around all day long is pretty taxing work. And we've got - You said 1 ,500 pounds. How big is one of those boxes? I'm familiar with a 55-gallon drum. I'm familiar with a pallet of goods that you might see in an Amazon warehouse. How big are we talking? Yeah, so think of this as the types of boxes that would flow to a retailer, right?

1:02:26To a big box retailer. Sure. And so they're unloading those from ocean containers onto carts that are about three and a half foot wide by about six and a half foot long. And so they just load up as many as they can on the cart and then take it to, you know, usually these are getting dispatched out to big box stores. And so, you know, there might be six tractor trailers that are getting loaded up to go to six different stores in a region. So we're working in the Sumner, Washington area. So I'm in Seattle port and helping basically get distribution out to Pacific Northwest. Can you talk about some of the differences about unloading at a port versus what Amazon's Kiva Systems does within the warehouse?

1:03:15And some of the different challenges that you face versus what Kiva is doing. I imagine that there's some learnings that cross over, right? Yeah, so the way you can think about logistics is there's inbound flow, you know, products coming from manufacturers around the world, a lot of it coming from China. and then that ends up in some distribution warehouse ready for people to buy. So it might end up in your local big box retailer. You can just go and buy a fan off the shelf or it ends up in an Amazon warehouse, an Amazon fulfillment center. So the inbound side of that is to unload the ocean containers and then bring it to someplace where it's being stored or bring it to a retail store or to an Amazon Fulfillment Center.

1:04:06And then an Amazon Fulfillment Center, yeah, that Kiva network, or now Amazon Robotics, what they call kind of the Hercules Drives, is their storage array. So Amazon will have multiple mezzanine decks of those Kiva pods full of all kinds of things that you might buy from Amazon. literally can have a million different SKUs in a building. And then when you order it, almost immediately, the system knows where, it has lots of those, and they're stowed across Amazon's network. It quickly calculates where's the most optimal place to deliver this to you. And then a robot goes and gets that shelf and brings it to a picker, brings it to someone who pulls it out of those shelves, puts it into a tote, and then those totes get routed to a pack station, gets packed, thrown, and then it gets sorted to a truck.

1:05:07And then usually, you know, either to FedEx or UPS or Amazon's delivery network, or to USPS, Amazon can kind of deliver into any of those outbound delivery networks. And so yeah, three phases coming in from the manufacturers stored and ready to be bought and then shipped to you how how uh where are we at in the kind of ai journey of these collaborative robots um i imagine that there's tons of work that can be done with just hard-coded business logic drive two feet forward take a left and it kind of just is almost like a conveyor belt on on independent wheels versus the far future where the robot has a brain and is just making completely independent decisions and decides where it goes and problem solves and reasons.

1:05:59And we're on the cusp of that. But I imagine that there's a journey that we're going through. And so walk me through where we are in terms of that journey. Yeah, so we've made a lot of progress from the days where you just like follow the tape line on the floor. Yep. Right? Now, robots can generally sense and perceive and navigate commercial environments autonomously quite well. Usually, you know, at human walking speeds, maybe a little faster, but the kind of self-driving problem is reasonably well solved in commercial spaces at those types of speeds. And that's generally done with a LIDAR. Oh, really?

1:06:43and maybe stereo depth cameras. And so LiDAR-based SLAM is how it localizes, and then navigation and planning. And that can be done in a way where it can detect humans, obstacles, navigate around things. And that's the capability our robot has, and it can do that in hospitals, in ultimately airports, stadiums, in and around people quite safely. That technology works. that what's coming now is you can talk to robots and the robots will make the the high level plan and instruct that um where we need to get to is what we can do with our hands right where you know like open up your airpod case and pull it's a very complicated set of motions that we do without thinking about it we don't quite have that capability yet got it talk to me about the lidar supply chain and cost structure in your business.

1:07:43It's been a controversial debate point for a long time in autonomous vehicles. But if the economic model works, it feels like it's just pure value add. Is the cost of LIDAR getting lower? Are you thinking about solid state LIDAR coming down the pipe or is it already available? Are you banking on a reduction in LIDAR costs over the long term Or does your business model, just by nature of how much value you're adding, you're fine paying 50K or something like that, if that's the number? I don't know. Yeah, LIDARs are in the kind of – they used to be 50K. They're in the kind of$3 ,000 to$6 ,000 range right now.

1:08:23Yeah. And so you're right. You do have to add enough value. You're not going to put that on your Roomba, right? But you can put that on an industrial robot and get a payback. Obviously, we'll want to keep seeing those costs continue to come down, but it's not a prohibitive element for a lot of the work that needs to get done. Can you talk about form factors broadly and what guided your thinking towards the proxy, the initial product, and other products in the suite? feels like you guys distinctly chose not to do humanoids, even though I'm sure various VCs thought, hey, why don't you have you guys thought about doing humanoids?

1:09:08Have you seen this demo? Have you seen this viral video? Yeah. So I'm curious, you know, kind of the decision making that went into the current and pending form factors. Yeah. So, I mean, as much as like the humanoid hype seems to have been peaked in the last, you know, 12, 18 months, I think Elon did a lot to kind of fan that. Humanoid robots have been around for quite a while. Agility systems have been, Aptronic, these guys have been at it for a while. And so when I was leading robotics for Amazon, we studied deeply humanoids. In 2018, I went through my hype phase on humanoids in 2018. I got really excited about them.

1:09:53What agility was showing at the time and legged mobility looked like it could actually work, and it does. And so I had my team at Amazon do a full analysis of everything. We weren't going to automate it another way where a humanoid could help. And I remember reviewing this paper. There were 40 different use cases where a humanoid could be great, right? And then we looked at all 40 use cases and we said, actually, to solve these problems, we don't need a humanoid. And in fact, a humanoid is kind of too complicated. You really want wheels. You kind of want to move more than three and a half miles an hour.

1:10:29And is the number of motors a potential issue, too, from a degradation standpoint? It's like, you know, we've talked to other robotics founders that say, you know, I'm using robotic arms in my facility. and we already have to replace those motors all the time. And then a humanoid might have an order of magnitude more and actually be less productive than some type of robotic arm. So the number of motors really does drive the overall cost. It also drives the complexity of the controls, right? And so you get into a world where you need AI-driven controls. But the problem, the real problem that people don't talk about very much with humanoids is getting strength out of rotational motion is very hard, right?

1:11:12you you're effectively because you're just doing a short throw you don't have the kind of momentum flywheel effect as you get the torques rolling on your electric vehicle right you're moving through very short distances so all the power is basically your electric magnet and then your rare earth magnets and and so you end up needing bigger and bigger motors to get that kind of power and so and then you want to wind them with the absolute highest density you can and so they end up and then you want to run them at almost the peak current that you can to get the most strength. So the problem is humanoid robots either have to put this way big motor that doesn't look right in the shoulder.

1:11:54But what they're typically doing is they're hand winding the motors, they're pushing the current to the max, and even then they're getting maybe 60 % the strength of humans. And they burn those motors out. So the motors are very expensive and they burn them out very quickly and they're still not as strong as a human and so it just it we need some breakthrough you know the pneumatic like you know boston dynamics had that atlas robot that like could do backflips and everything pneumatic is 10x the uh the power of an electric right and but you can't you can't really make that system reliable in production so do you think there You think there are more consumer use cases for the humanoid form factor, maybe around the home?

1:12:41How do you think about applications outside of industrials? You know, I have struggled to find someone. Someone mentioned one to me the other day that seemed great, which is like is walking your dog. I think humanoid walking the dog would be would be quite interesting, quite cool. I suppose you could have a quadruple. Man's best friend's best friend. but uh um otherwise i am i am not bullish i do think there's some some cool robots recently that are more kind of friendly look like they're kind of playing a game with a kid like i i think kind of the emotional companion idea is quite interesting um but yeah getting the strength is tough i mean even just thinking about the human arm like the force that's generated from the human arm is from like the bicep muscle, which is much bigger than the actual joint.

1:13:31And so if you put a motor on that joint, you're not using this. We need a humanoid with just absolute cannons. There is a humanoid company that's trying to create the muscle fibers and pull that. And that sounds like some of the pneumatic projects. And maybe it'll be a hybrid. In terms of training and AI development, there's been this talk about the sim to real gap. I don't know how closely you've been tracking this, but obviously generating data for robotics has been very difficult, but now there's this new paper that Semi-Analysis was talking about yesterday, all about training in simulation, basically Unreal Engine, you build the robot virtually, you have it walk around, learn as much as it can in thousands of years of artificial data, then there's going to be a gap between what it experiences in simulation and reality.

1:14:18And so what you do is you take what it's learned in simulation, and you run that on a robot in a basically wired up with a power cable so that it can run forever. And it tries to do the moves that it learned in simulation. It messes it up, but then that generates more data that feeds back in. Does that seem like power generation, all the mechanical issues aside, does that seem like an interesting path to go down for actually solving the algorithmic and like the AI piece of understanding how these robots will actually choose what motors to move at what times. And do you have any experience generating data?

1:15:00Just kidding. I'm just kidding. Former CTO of Scale AI, if anyone's listening that's not familiar.

1:15:08No, that's – so the challenge in robotics first is how do you get – I mean, it's the same in large language. Well, what's the pre-training phase, right? How do we get some base level in pre-training large language models? It's to kind of understand how words are likely to follow each other, right? Just statistically. So motor actions, what's likely to, you know, to cause the arm to move forward and things like that. But the hard part in any AI system are the edge cases at the end when you're interfacing with the real world, right? And, you know, fortunately, we have all this large language model data.

1:15:53We have all this data from the Internet to give us reference examples of what the real world of language looks like. Right. And so we refine on that. And then we use human preference to refine even further. And that's how we get, you know, ChatGPT. in the robotics world, data from simulation, data from multiple robots, data from teleoperation, all of these are kind of techniques people are using to feed some data into, you know, what's kind of the pre-trained base model that gets some statistical correlation. But when it comes to learning the edge cases, right, when it comes to, hey, that doorknob is higher than this other doorknob or the doorknob turns upward instead of downward.

1:16:41You and I actually self-play to figure that out. We come up to a doorknob that doesn't. It's funny. We have a door at Cobot that you can either push the handlebar or there's a handle. Well, everyone tries to push the handle and then the door doesn't open and you have to push the. And so humans get confused, too. And we do this kind of refined self-play. And I think right now we're very much focused on the pre-training phase, just how do we get enough data to have something that roughly moves its hand toward the door. To really solve this problem, though, we've got to learn how to self-play in the real world like you or I do because there's all kinds of novel stuff we're going to run into solving real problems.

1:17:24Well, good luck with that. That sounds like an easy task, but I'm sure you're up to the task. And it's been fantastic talking to you. Yeah, this has been super insightful. Yeah, I mean, it's such an exciting industry because it's really just like we're still just on the early part of the S-curve. And there's going to be fantastic advancements. So good luck. The future is going to be amazing. Awesome. Thank you guys. Appreciate you coming on. We'll talk to you soon. Thanks, Brad. Thanks so much. Talk to you soon. Next up, we have Kian from Nucleus coming on with a big announcement. something like 10 years in the making close to it maybe seven years uh we'll bring key on in let's play some soundboard how you doing welcome to the show that's a great intro the tweets are flying oh my god you guys seen this yes you're seeing this seeing this break it down for us explain what's happening there's nothing like a launch day i'm trying to figure out guys is this is the gattaca or is it theranos because people can't They can't make up their mind.

1:18:25Oh, yeah. We're going to find out. They're trying to figure it out. They're trying to figure it out. What's going on? Let's give some context to the audience. Nucleus has launched Nucleus Embryo, the world's first genetic optimization software. Basically, parents can give their children the best start in life. They can pick their embryo based off of physical characteristics like eye color, IQ. They can go to disease risk like cancers or heart disease. Basically, we really believe parents can get all the information that exists about their embryos, and they can pick however they want. For me personally, it's been 10 years in the making.

1:18:55The journalist actually covered it today in the Wall Street Journal was a journalist that covered my gene editing in a warehouse in Brooklyn 10 years ago. Yes. Wow. Overnight success. You know, it's a long time in genetics. Yeah. So break down the state of the art because like embryo screening exists. I think most parents in America, at least if they have the means, do some sort of screening while the embryo is growing. Is this purely for IVF? Is this just going a layer deeper? And then I want to talk about the regulatory and FDA component as well. Let's talk about it. So basically, if you go to an IVF clinic today, you're a couple.

1:19:32The vast, vast, vast majority of clinics. The first thing I should understand is that the IVF process is principally controlled today by clinicians or doctors. Honestly, couples don't have as much liberty in our perspective as they should. It's their baby. It's their embryos. They should have the right to that information, and they should pick off any vertical. However, today in the clinic, what generally happens is people test embryos for very rare and severe genetic conditions. For example, like a chromosomal abnormality, like a Down syndrome, for example, or even a condition like cystic fibrosis or Tay-Sachs or PKU, right?

1:20:02These are conditions that are very rare, that maybe someone might have a carrier for cystic fibrosis, but again, it's pretty rare. Then there are conditions that we've all heard about, things like breast cancer, things like coronary artery disease, the things that actually kill the vast majority of people today, right? Chronic conditions kill the vascular people today. Those conditions are just not tested for in the clinic, even though we have very good science, actually, that can make those predictions. How do we know this as a DNA company? Well, that's what we do, right? We build models that predict disease, and the way you test those models is in adults.

1:20:30So we go from adults to embryos is actually because we can basically well validate these models to show that they work in both the embryonic context and in the adult context. And so what we're really doing is we're going from, okay, instead of just looking for really severe, like, down-systric cystic fibrosis, why not do breast cancer? Why not do heart disease? Why not do colorectal cancer? Why not do schizophrenia? Why do Parkinson's? But then why stop there? And this is really the important thing because ultimately, if you think about disease and traits, the extreme version of any trait is actually a disease.

1:20:59Height is a good example of this. One extreme end is like John, for example. He's like Markman syndrome. Then the other end is like me, dwarfism, right? It's like disease on both ends. So IQ is another example of this. One end is like autism. them, the other end, it can actually be some sort of cognitive, basically, challenge that people have. And so when you think about it, we start realizing that people have drawn a line in the sand saying you can get rare diseases, you can get common diseases, but then they really say you can't get any traits like height, even though the best predictor we have today actually in the world, the best polygenic predictor is for height.

1:21:34So as a company, we've completely reimagined this and said, wait a second, what's going on here? You should have access to the entire stack. rare diseases we do, cystic fibrosis, common diseases like breast cancer, and also traits, all the way up to something like IQ. Yeah. So, I mean, that test, are you just giving people the data? Because I imagine that once you get into particular recommendations, that's more of what I would expect a licensed doctor to need to do. Well, yeah. My sense is that you can allow people to get the data from their doctor and then feed it into Nucleus. Is that correct?

1:22:09So that's, that is correct. And actually we have a couple, there was like 10 announcements today. You know how we do it. We like to do 10 announcements in one day. We are actually very, very excited to announce a huge partnership with genomic predictions. So genomic prediction is actually the oldest embryo testing company that exists. They've done genome-wide tests in embryos for almost a decade at this point. And I think they've done over 120 ,000 couples for PGTA, which is a specific kind of test. And so we're actually partnering with them. So we make it very easy for genomic prediction customers to request their files and actually port it over to Nucleus.

1:22:39But really, this isn't just for genomic prediction customers. Anyone who's undergoing IVF can go to their clinic and say, I want my embryo's data. You can take that data. You can upload it to Nucleus. And then all of a sudden, you know, the application of DNA makes this technology universally accessible. Now, how much of the benefit is actual algorithmic analysis bringing in other data points to contextualize the data versus just better UI and better hydration of existing text? Because we had a friend on the show who was talking about getting some medical results from a doctor. the doctor's office was closed.

1:23:20It took two days until the doctor was going to be able to interpret the results. He was able to just take a photo, upload it to ChatGPT and say, hey, is this really, really bad? Should I be panicking? Because it seems somewhat out of the range. And ChatGPT was able to say, hey, you still got to talk to the doctor, but this isn't the craziest thing I've ever seen. This isn't way out of distribution. And so that's almost like a pure UI layer, but extremely valuable. I know it might not be like the right narrative for some people that that it's not as innovative, but I think that all that matters at the end of the day is giving people benefits.

1:23:52It's always both. It's always both. Fundamentally, technology, just for technology's sake, is not siliconized about. It's siliconized about making something that people want, and that people can actually use. So you think about the nucleus innovation, it's two-pronged. One is in the informatics. I've been doing this for five years. I would argue to myself that I probably spend too much time developing the science. Science in a nutshell isn't actually very useful. You need to expand access to it. So on that point, we do multiple different kinds of analyses that make it such that we can actually provide the most comprehensive analyses that exist today.

1:24:25But moreover, and this is really, I think, a key point to your point, John, is people understand them. People can see them. I mean, you can pull up the platform. I'm not sure if you guys have shown it already, but it's very easy to sort, compare your embryos. You can actually name your embryos. You can stack rank your embryos. You can understand what the score means. We lead with overall risk. Or we tell you, for example, instead of saying you're in the 99 percentile for genetic risk for a condition, which, you know, what does it actually mean? We say, hey, you have a 5 percent chance or the like of, let's say, schizophrenia or some other condition.

1:24:52In other words, by leaving overall risk, people have a much greater intuitive understanding of the results we're communicating to them. We have genetic counselors on hand. So this really is a – what are we showing here? Are we showing something? Are we showing the – Yeah, yeah, yeah. We pulled up your website. That's another thing. That's a fun one. That's an Easter egg. That's an Easter egg. That's the kind of approach that we're taking here. And I think consumers are responding to it, right? Like people want to have access to their data. The clinician, the doctor shouldn't decide what embryo implants.

1:25:18You should. Okay. So talk to me about what requires FDA approval. Obviously, new medical devices. Like if you were developing a machine to take in an embryo and sequence the DNA, I would expect that the FDA would want an approval for that medical device. But if you are taking data and just showing it to a customer in a different UI, that feels like probably a very light FDA process. And then there's probably a continuum in the middle where once you're making a recommendation, they have rules around that, right? We as a company do not tell you which embryo to implant. You know, basically parents, the couple has complete agency to decide how they want to use the information to implant their embryo.

1:25:59Moreover, let's be clear, height, right? I mean, can a height analysis be a medical device? You know, it doesn't even make sense, right? IQ, height, there's traits, for example, we all, You know, traits are something that I don't think actually belongs in even the kind of infrastructure thing about medical care, right? These are things that go beyond medical care. These are things like, you know, that people just kind of intuitively know and that there are DNA tests done every single day due to see for these analyses because they're not disease analyses, right? So we do both diseases and traits to be clear.

1:26:26My point is many of these innovations, you have to wonder like, you know, should the government say if someone can or cannot pick their embryo based off height? That doesn't seem right to me. I think it should be in the complete liberty of the individual to decide that. Yeah, but I mean, we're a democratic country. And so if a huge swath of the population says that the FDA should review that type of test or that type of analysis, it could happen. I mean, the FDA reviews all sorts of different stuff. And so I guess the question shifts to like, do you expect a change from FDA on the way these analysis tools are regulated?

1:27:08I think right now the most important thing is just putting these high quality rigorous scientific results in people's hands and helping them basically have healthier children, helping them give their child the best start in life. I think that generally speaking that people should have more liberty, more choice in medicine. I think the broader longevity trend actually touches on that point as well. So that's what we're excited to do at Nucleus. Yeah, I mean, the fact that you're partnering with a company on the actual medical device side, they are doing the sequencing of the embryos. That really takes it out of the Theranos question entirely in my mind.

1:27:44I feel like you should be beating the drum there a little bit more. It's like we didn't say we created some new device, but I don't know. We ship. That's the difference. We ship. It's live and thought, baby. Don't look at it. Go use it. That's the evidence. Why is there a pudding, okay? I love the visual of John and his wife selecting between embryos and it's like 610 or 72. Tough choice. Well, if we go with the 610, he has, you know, 30 feet. He could potentially fly commercial once in his life. We can actually play this game right now. Okay. Here, we're going to play a game right now. I'm going to put it in the chat.

1:28:21Pick your embryo.com. Okay. Everyone listening to this. Pick your embryo.com. I'm going to go to it. Oh my God. Here we go. Little Easter egg here. Okay, let's see what's more important to you, John. Intelligence or muscle strength? Come on. Oh, absolutely. Muscle strength. Let's go. The future is bodybuilding. Let's go. Okay, height or lifespan? John would take a – he would happily have a 5 '2 sun if he had, you know, top 0.01 % bodybuilding genetics. Exactly, yeah. Okay, so which one here, lifespan or height? Come on, lifespan. Lifespan, let's go. Let's go. Let's go. Let's go. Maybe low depression.

1:28:56You got to be golden retriever mode. You need low depression. Let's go low OCD. I don't mind bouncing around a bunch. Risk taking anxiety. Let's go high risk taking. There we go. Let's see what you've got. Is this some generally stuff going on? This is great. Nadia. I got Nadia too. The enduring athlete. Let's go. Physically strong, cautious, built to last. This is great. Is this driving a lot of attention, a lot of downloads? Is this going viral yet? It seems like something that's designed to be shareable. I think we just dropped it right now. Technology Brothers, we got you the exclusive. Let's go.

1:29:32Let's put it out there. You can pick your embryo. People say, what's it like? Maybe you're not doing IVF yet. No problem. Funny, only 9 % of people choose Nadia. Okay. Well, we're contrarian. We like that here. Yeah. That's fun. It's great. Oh, well, congratulations on the news. Congratulations on the launch. Yeah, the pace is wild. Last thing, what's going on with, have you seen these Just Blood billboards? Oh, yeah. They're all over LA. So there's someone who's running a campaign right now, justice for Elizabeth Holmes, claiming that Theranos was not the scam people think it was. And there's a documentary coming out and there's billboards all over LA for just blood.

1:30:13Like it's just blood. It's not that big of a deal. And John, just to be clear, there's an exclusive on Technology Brothers next week about from this person, right? They're going to tell their story next week just to make sure you invited them already. I hope. I want to hear the story. We are toying with the idea that someone reached out to kind of connect us. We're thinking about doing it, but we're not 100 % sure that it would be appropriate for the show. Based on the website, I don't know if it's appropriate. Yeah, it doesn't look like it was designed with Figma, so I don't know. We can't quite do it.

1:30:42It's a little bit. The team definitely doesn't use linear. Yeah, but they claim that Elizabeth Holmes has been proven innocent, and so it's a bold claim. We like to see people making bold claims. By what jury is my question. The jury of someone who knows HTML. Keon. Always a great time. The energy is off the charts. Fantastic. Electric. Electric. Thank you for coming on, firing us up. Congratulations on the launch. We will talk to you soon. Talk soon. I'll see you on Twitter for sure, okay? Yeah, we'll see you there. Bye, guys. Bye. He's going through launch day right now, which is just like 40 notifications every minute forever.

1:31:21I love it. Well, next up, we have Kathleen from Valthos coming into the studio. Welcome to the stream. How are you doing, Kathleen? Nice to meet you. Good. Thanks for having me on, guys. Yeah, it's great to have you. Would you mind kicking it off with just a little bit of an intro for those who might not know? Yeah, absolutely. So I have been at Palantir for the last seven years. I built our life science practice there. So taking some of the same platforms that Palantir uses in defense and intelligence and a bunch of industries and then bringing those primitives over to pharma and biotech and researchers who need to study biomedical data in a secure and collaborative way.

1:31:53So really building out end-to-end drug development workflows all the way from early discovery. But then I left a couple months ago, and now I'm working on something new. Very cool. So yeah, happy to chat with you guys. Yeah. Awesome. I wanted to get your immediate reaction and help us kind of contextualize the news that came out yesterday or the day before, which was that two Chinese nationals have been charged by U.S. federal authorities with conspiracy and smuggling after attempting to bring a dangerous biological pathogen, which I'm going to botch the name, into the United States. This fungus is classified in scientific literature as a potential agro-terrorism weapon due to its ability to devastate key crops such as wheat, barley, corn, and rice, causing a number of issues there.

1:32:39So I wanted to kind of get a, you know, you don't have to go into too much detail, but kind of a high level background on bioterrorism broadly, kind of maybe some like prominent examples, and then even just get your immediate reaction to the news, you know, is this kind of thing surprising to you? Or are you surprised that we're not hearing more headlines about it all the time? Yeah, totally. So I think the to the second question first, it's like a bad surprise, obviously, but we are absolutely entering this era of an elevated risk of biothreats. So sadly, I don't think it's something that we should be that surprised about.

1:33:19I think this story in particular, and we don't know that many details on the story, so I won't speak too much to it, but the one it highlights how easy it is really to have any biological material pass across borders, that obviously both natural or unnatural. And then the second is that any kind of agricultural pathogen is also an enormous threat. So I think when people think biodefense and bioterrorism, they think anthrax or smallpox, which is obviously horrible. But the idea that you can introduce a pathogen that would devastate one of our primary crops, that would have massive health impact.

1:33:53But it would also completely destabilize our economy and send us into something far worse than what we would see with COVID. So I do think when we in governments think about biodefense, it's very much both in terms of human health and agricultural health. But to your other question on what's the primer on biodefense, what are we worried about? I think there's always been this. So we've always had the problem that people study the most pathogenic organisms in the world. They usually do it in a biosafety lab, a BSL lab. These are all over the world. Some regions are obviously more secure than others.

1:34:28So there's always this threat of something natural leaking or being used maliciously. But basically two things changed recently. It made that a lot worse. One is that our ability to edit genomes and actually start changing those pathogens, there's a bunch of new tools to do that. So things like CRISPR, which you hear about usually in the medical sense of editing a genome, are available on pathogens. Synthesizing new nucleotides and new DNA to do that with, also now commercially available. We could order nucleotides or you could print them out in some cases in a desktop printer. So that made it, now we're dealing with things that nature has never seen.

1:35:03And then in the last couple of years with large language models or biospecific language models, we have this idea of AI uplift. So it means that someone like us, not to underestimate your biological skills, but could actually be coached into how do we use these tools guided all the way into making something that is actually way worse than anything nature has ever seen. And this isn't like science fiction. Like Anthropik talked last week, I think, about Claude IV. they did internal safety trials they noticed an uplift that goes way beyond just oh it's easier to google a paper about how to do this and really into that um allowing a novice to access something dangerous so they released cloud four with you know new security standards to try to combat that but a lot of models aren't like that not everyone has those security standards you know yeah i mean this this has happened like uh was timothy mcveigh looked up how to build the bomb with fertilizer and basically blew up a massive government building.

1:36:01And the wreckage from that was insane. If you think about what it would take a terrorist to work in a biosafety lab, it's obviously very complex, but getting easier as you can have an LLM coach you through the process essentially, right? Is that roughly the nature of the threat? Yeah, both in terms of the actual steps to take and then what to change about a virus that would make it either evade some kind of countermeasure, which is like some kind of medicine that we have for it, or be more infectious. And I think the real, like, what has a lot of attention and weighs on people's minds is that we're not talking about, like, a state-sponsored nuclear program, that you need these, like, massive budgets and facilities to make something like that.

1:36:44Like, it's a laboratory and a computer. So it's a really different dynamic of threat than what we've been dealing with in the past. And it kind of self-replicates by itself. That's, like, the whole goal. It's an interesting story to me, because obviously over the weekend there was the Ukraine story around Operation Spiderweb, and that was relatively asymmetric in that, you know, whatever the cost of the operation, the trucks and the drones, just, you know, even if it was$100 million, which seems really, really high, right? Even if it was taking, you know, huge teams, it destroyed, you know, a billion dollars plus of assets on the other side.

1:37:20And then when you talk about bio-warfare and bioterrorism, it's like, okay, one or two people with access to a lab could potentially do billions and billions of dollars of damage. Do you think this is a wake-up call or this will be a wake-up call for the government broadly? How do you expect, you know, what are the different kind of ways in which the United States can defend itself from these types of, you know, obviously this wasn't whatever's being reported isn't a direct tack, but a potential threat in the future. Yeah. I mean, the good news is DOD here, MOD in the UK, there's a lot of defense organizations already thinking about this and really trying to get ahead of it.

1:38:03It means that we have way more to do, but at least it's on the agenda. And I think it comes down to, like, the best defense would be to prevent this from happening. So putting better safeguards on models, putting really strong regulation on synthesis. So who can synthesize what? How do we track that? That's great. There's actually some legislation in the works on both of those things. But that works domestically and it works to contain an accident. But if we're actually talking about international collaboration, like those kinds of regulations are not really enough. So the way then we get into thinking, how do you deter something like this, to your point?

1:38:40Like, what is the defense against this? And there's really like three pillars that go into it. The first is how quickly can you detect that something happened? So in this case or in future cases, do we immediately know that something new is circulating, that it's high risk to us, that it maybe will evade any kind of countermeasure that we have? And can we know that before it's an outbreak and we're sampling from a hospital? and then that leads into the second one which is how fast can we design or update a countermeasure so like an antibody or some kind of biologic to combat that um and that is really like if we can diagnose and develop as fast as possible then these weapons are much less powerful because you're not talking about billions of damages you're talking about a couple cases that are quickly contained and the last part of course is attribution so if you can actually say where did this come from?

1:39:33Did it come from nature? Did it come from engineering? Did it come from nation state? That lets you bring the rest of the DoD and the State Department and our allies towards preventing something in the future. And if you can get that cycle down to hours, really, in terms of detecting, stopping, and then attributing, then you actually have a really robust profile for defense. And there's a bunch of new tech that's going to help make that better. So this is scary, but it also is, there's good news on the horizon too. Can you talk a little bit about the bio practice at Palantir? I mean, there's been a lot of potentially like misunderstanding or misinformation about how Palantir works.

1:40:13You know, I think most people who understand the company at this point understand that it is an ontology platform that's on top of a large database. But I think what I'm struggling with is I understand if you're using like the Airbus case study, you have a database with all different parts of the airplane and then Palantir understands helps you understand how the different pieces and lead times for you know this screw and this seat belt and this engine part fits all together so if you're demand planning or figuring out how to manufacture airplanes that's a very helpful tool to understand your supply chain that makes sense to me in the very concrete widgets business that is airplane manufacturing although it is obviously very complex in the bio or pharmaceutical space i don't really understand the nature of how large these data sets are are we talking about trial data or manufacturing?

1:41:01Is it all of these above? How does all that fit together when you're thinking about applications of understanding large data sets in just the bio world broadly? Yeah, definitely. Some of it is more similar to what you described with Airbus, where we're talking about manufacturing. We're talking about something that's like, it's a process with a lot of moving parts how do you make these all synchronous and update uh when you need to there's also obviously biologics manufacturing um some of it is more in the logistics end which palantir also talks about quite a bit so when you're running a trial making sure that patients and the medicine that they need are in the right place at the right time with the right support staff that's actually looks pretty similar to coordinating flight routes or staffing a hospital um the part that so those parts are super similar to the rest of palantir The part that's probably most unique is when we're actually talking about that trial data or talking about patient level information.

1:41:58And there it really gets back to some of the core concepts of Palantir is how do you work with multimodal data and see patterns across it? So if we're looking back historically on all trials that we've run and we want to start trying to identify what kind of patients have the best response to this or what are potential side effects? can you start linking together data that's from like a medical record with samples from a lab with sequencing data if you have it and can you do that in a way that is completely secure completely audible and complies with all the regulation in the space and that's really the the niche that palantir maybe it's not always well understood to put into but uh that is what that platform was built for can you talk a little bit more about uh the i don't know like the long-term future of what you're building.

1:42:46I know that you can't go into it too much, but

1:42:51there are a bunch of different vectors and opportunities around what we're building in bio and what we're trying to prevent. There's a little bit of game theory going on here. So what would you like to see the United States really, really dominate going forward? And where are the biggest opportunities to both increase biosecurity and then also help accelerate the developments that we need, the good stuff. And that is what I'm most excited about. All of these technologies that are really scary when we're talking about them like this actually do have the potential to give us this enormous global advantage in our bioeconomy and how we respond to these threats and make medicine too.

1:43:33It's two sides of the same coin. I think the two areas that we're most excited about. One is on that detection pillar. So right now, a lot of the ways that we understand what's going on around us is super analog. We have like a list of pathogens we're looking for. We test and we get a yes or no if those exist and not much more than that. We've already seen a shift in this space towards sequencing. So actually getting the DNA sequence from anything that's in the environment or any of the sample. So that both broadens the scope of what we're looking at rather than just having tunnel vision on the pathogens we know about, which means that we can start detecting unknown unknowns and things that we've never seen and make a risk assessment.

1:44:14Also using sequence data lets you have this much deeper level of insight into what the risk is rather than just there's a virus here. You could say these mutations make it more adapted to this type of host or make it more dangerous or make it evade a certain type of medicine that we already have. And then you can take all that information that you get on this more robust detection layer and use it to drive countermeasure design. So we now have like every biotech that you guys talk to probably talks about programmable therapeutics, where we're moving to this era where you can update based on how the targets update.

1:44:49And if we have this deep level of intelligence, we can also start thinking about rather than have a stale stockpile of medicines that were made 10 years ago, can we actually see a threat immediately change the countermeasure and then start deploying that immediately. So I think getting that cycle down really tight, that's what the future is hopefully. Are we testing enough? I know when I go through TSA at LAX, someone who was doing research, maybe for the CDC I talked to, was saying that like they basically just focus on LAX because that's like the biggest hub of virus. It's gonna freak everyone out who flies.

1:45:28It's like, if you're gonna get sick, like if you go to LAX, that's where it all starts. It's like a petri dish. I know that they swab my hands for, I think it's like bomb making materials, but should they be swabbing my hands for new pathogens? Should we be doing more in the detection? Should we sequencing random farmland to see if there's new invasive pathogens that could be targeting our crops? And then like, how do we even pay for that? Is that something that the government should put the bill or corporations should be incentivized to pay for? What is the actual upping the amount of data that we're ingesting look like?

1:46:05Yeah. I mean, I'm always going to say more data is better. Yeah. But the question is like, yeah, how do we get more data? How do we incentivize more data, buy more data, test for more data? Yeah, absolutely. So I think of course we should collect more. The type of sequencing actually helps drive down the cost because you can target a wider range of things you might worry about rather than setting up individualized programs for specific pathogens when maybe that's not a threat. But I actually think with the caveat of more data is always better, we actually do collect a lot of this data today. You don't necessarily see it because some of it is in wastewater or environmental samples, like you're saying.

1:46:45We don't always extract that much intelligence from the data that we do collect. So we know that a sequence exists. We might not necessarily know what that means in terms of health or the impact of that variant or that mutation. So I do think there is benefit and there's cost-effective benefit for collecting more. But a big piece is just on of the data that does come in, how do we build the right models and the right software to interpret that? Makes sense. Makes a lot of sense. I think that covers it for now. I would love to have you come back on when you're ready to talk more about specifics on what you're building.

1:47:22And I feel grateful that you and the team are doing what you're doing thank you and just work a little bit faster please um but uh thank you for coming on and giving us some insight here and uh congrats on uh on starting the founder path thanks very much guys great chatting and hope to chat soon cheers thank you next up we have roy from clueley coming back for an update he's hired 50 interns i think or something close to it he said they're bringing every intern on. They're bringing every intern on. We got every intern coming in. Well, welcome to the studio, Roy. How are you doing? Oh, let's go.

1:48:03There they are. I think we're overpowering you. Can you hear us? Yeah, yeah, we can hear you. Yeah, make sure we're zoomed out all the way so we can see everybody. We got a small army here. This is incredible. How big is the team? Kick us off. How many you got at this point? The team is 11 full-time plus the interns. How many interns you got so far? Interns, bro, we're closing in on 50, brother.

1:48:30That's amazing. Congratulations. What are they all doing? How do you manage everything? Is it purely social media? Is that what you want them to focus on? Growth? Yeah, yeah. Growth marketing. The only goal of the company is get 1 billion eyeballs onto Cluely. So we have unrestricted creative freedom and permission to do anything and everything just make the company go viral. Every single person you see behind you has over 100 ,000 followers on some social media platform. Wow. Most of them are 200 ,000 plus. That's remarkable. Me too now. Me too now. Yeah? There we go. Oh, yeah, you probably popped.

1:49:07What's working? What platforms have actually been driving the most growth? I mean, I'm sure you've run a lot of tests. What have you learned that you can share? Bro, Ben, take it away, bro. Let's hear it. UGC has been really good. We just hit 10 million views today. 10 million views. Eight days. Wow. There we go. Hoping to get 100 million views in the next month. What platforms specifically are the most fertile ground for targeting your specific customer? Because you can imagine that there's a lot of folks who are AI curious on X, but then there's a much broader, more viral audience, more general audience on platforms like TikTok, YouTube, Instagram.

1:49:44What's working and what is the next, next platform that you're going to be focused on? yeah well we're trying to go viral on every platform regardless um but the main thing right now is instagram reels oh instagram reels interesting and what is the main value prop that you're hitting people with is it still the cheat on test thing or have you evolved at all what still actually interviews yeah interviews okay and uh has there been it this was controversial when you launched it is it still controversial in the comments are you getting flamed has anyone big dunked on you and has that driven virality?

1:50:19Is that actually a net positive? Instagram is not like Twitter. You could post the craziest shit on Instagram and they still will not think it's controversial. So how to make it controversial, we have to engage in bait some other way. It's hitting tool is controversial on Twitter, but on Instagram, you could have a white guy say the N word 10 times and it's still not controversial. You need crazy shit on Instagram. That's what we crack. Every single person here has very great viral sense. And watch the reels that do go viral. You see there's like ways that we've engagement baited the videos. And this is what we'll keep doing to probably a billion views a month.

1:50:51How long does it take to figure out if an intern is cracked? Is it like an hour or two hours? How much time do you need? For me personally, me personally, probably like 10 minutes. But for anybody watching probably would take like one or two weeks. There we go. There we go. How do you guys think about how do you guys think about product marketing? Obviously, you're just going viral everywhere getting all this attention how do you make sure that it that it uh while he's shaking his head doesn't think about it's not about the product it's about the attention go viral yeah um yeah but but but how do you the side of the street you know you uh make some ugc videos make some twitter posts you know you can sell anything you know in 2025 product doesn't matter you know i could jack off off the side of a building sell some videos of it for 20 bucks each make two trillion dollars it's crazy two trillion that that's intense um how do you guys think about burn uh is it on your mind is it on your mind at all i don't know if you saw the last tweet but as of literally like two days ago we're still we're still cash flow positive we're still fucking we're still wow let's give it up for the property let's hear it let's uh it sounds uh so you're charging for the product and people are paying are they at all satisfied or do they feel uh Like they got scared, bro.

1:52:05Like the product works. You're either using this as a consumer and it's working because like like you're passing your interviews and or if it doesn't work, you're not going to complain to me because I'm going to go right to your employer and tell them, yo, guess who's complaining about using the product? Like I'll get you blacklisted if you complain really.

1:52:22Where how are you thinking about how do you how are you guys thinking about product evolutions? What do you want to add to the product? Obviously, you want to help people cheat on everything. Where are you going to help people cheat next? We don't care about like the product is going to be led by the virality of the content. We have video ideas right now that we're going to try to push for different use cases. We're going to see which ones go consistently the most viral. If you can make something go more viral, then you can just build the technology after you have all the attention. So we'll figure out the exact use cases and exact niches we're going to quintuple down on once these guys get to work.

1:52:56What formats on Instagram Reels are like the most modern in terms of consistently viral? Like you mentioned like man on the street interviews, what do you do for a living? That's always been a fertile ground. What about, I see a lot of those like mobile game ads that look like, you know, you're fighting down some sort of bridge and then you go into the game, it's actually just a match three. What are the different formats that you like to pull from? every week there's two new ones and at any point there's probably 10 to 20 viral trends that is happening and these cycles so quick you need to keep your finger on the pulse these things will like expire immediately you need to be on the ball and like if i if i told you right now by the time people watch this on youtube like it would have all been expired well then we're live so so give us give us the latest and greatest like what's going viral today well right now we got 10 million views using a snapchat format okay viral for like the last three years to be honest okay and and i think that like we just have to get people who continuously scroll tiktok like six hours a day yeah but what's the what's the actual format that you use like describe the video what is the hook like break it down for me like you're explaining the like the art behind the viral format there's a caption it starts with the face usually a handsome dude or a pretty girl you're saying damn this interview starting with the interview was starting with the hard questions i should have been a cs major not a business major that's engaging because people are saying like bro like cs is way harder than business then it turns around the interviewer asks like like hey how are you doing why should we hire you and then this guy uses clue to generate a response but he can't fucking read the response so he reads it hella autistically like oh i revel in detail and and And then that is like another conversation point.

1:54:40Like people are cooking on the guy because he can't read properly. The guy is like doing a really dumb interview using QA. That's great. How are you guys using AI generated content internally? I know a lot of these videos that you guys are creating are just typical social media, vertical video. Do you have an intern that's just generating basically copy and pasting, making a bunch of other interns? B-roll, VO3, are any of these tools relevant? Anything clicking? Not yet. I think there's still like a 10 % left before they cross the uncanny valley. And the biggest thing is that people need to think your video is real.

1:55:14That is the difference between 100k views and 10 million views if people think of this real. Absolutely AI CEO is bearish on AI.

1:55:25Google needs like 10 more Chinese researchers to like figure it out. And once they push out the latest update, then VO3 will be there. But right now we need real people. Yeah, yeah, yeah. Well, I mean, what about just using AI as like stock footage replacement, not as the lead in for the video, not the entire video, but just like sprinkled in to illustrate a point, you know, an establishing shot of like a building, a helicopter pulling into a building like that. That historically has been kind of something that you would reach to, you know, Adobe stock video for VO3 feels like it's there, but are you not drawing on that at all yet?

1:56:03If there's a viral format when we need it, maybe we'll use it. But right now, it's really brain dead to go viral on Instagram. The formats are not hard. You don't need a helicopter. You need a guy, a camera, a really shitty camera, and you need a computer. I mean, what about those kind of AI mashups like Harry Potter Balenciaga or the kangaroo with the plane ticket getting on the plane? AI content can go viral when it's inspired almost by a human. It's not entirely AI generated, but it's using the tools effectively to create something that's still catchy. Do you think you'll be using any of that anytime soon?

1:56:40Probably very soon. We're scaling up. What you see right now is probably about less than 1 % of what the size will be by the end of this year. We are profitable. We're not trying to be profitable. We just keep making so much money, we can't help it. So we're really scaling this shit up. I'm not even trolling you. 1 ,000 creators are going to be shipping out content. We're doing a complete internet takeover. I love it. So why in-house? Why do they even have to be employees? Couldn't you turn this into a multi-level marketing scheme or something, a pyramid scheme? We're going to do this. Oh, that's what you're going to do.

1:57:11Okay. MLM. MLM. MLM. I love it. Are you guys worried that you could be infiltrated by journalists? I'm sure they're circling the house right now. The hit pieces are going to come. We're doing a softball interview right now. I mean, the person that's brave enough to try to do a hit piece on the Clulee Army is going to be. I bet they're dying too. Look, more eyeballs is better. There's no companies that ever died from a founder being too controversial. You got deal fucking infiltrating with genuine spies and they're still doing fine, bro. Worker 17 guys, they're still kicking. Like no company ever dies from being too controversial.

1:57:49You die because you don't make enough fucking money. Yeah, yeah, yeah, yeah. Speaking of making money, what's the pricing model right now? Are you doing anything on price discrimination? Is there a super high tier if you get a whale? What does it clearly whale look like? Can I spend$2 ,000 a month on this service? Yeah, you should add a tipping feature too. Yeah. People should be able to tip you guys if they have a good experience. You should add really financialization, pay as you go, high interest rate loans. Just really push it. Make it sports gambling in there maybe. Just throw it all in. Yeah, I mean it's$20 a month for a consumer,$100 a year.

1:58:22And our top line revenue is really being driven up by enterprise. You're going to have to talk to the sales team to get a custom pull. But you know, like, there's a lot of - Wait, are you serious? Is that more on the sales side? Who are the enterprise? So you sell the SDRs? You guys laugh because you think I can't sell enterprise because I'm - No, I don't believe it. I trust. So serious, like these Fortune 500 CEOs, like these are like 35 year old dudes who sit there scrolling through Twitter laughing at my posts. Yeah, yeah, yeah. Yeah, no, no, it seems legit. It makes sense. No, I believe it.

1:58:54But I mean, you're not going even higher tier? Like what's the$2 ,000 a month clearly vision? For a consumer, there's a lot more we can do with more compute. But right now we're like, to be honest, I didn't expect to grow this fast. The eng team is quite small. I'm like spending a lot of time trying to hire more competent engineers. We have a lot of backlog tasks that we need to fill out, especially for this last contract that we signed. So we're full-time focusing on the one big guy that we got right now. And after that, then we'll try and scale this up. But right now we're focused on the one big client that we signed.

1:59:25Yeah. Talk about your compensation strategy that people want to know. You said you can raise infinite capital and you're so confident. I believe you. But I'm curious to get some more insight there. Bro, I feel like it's so retarded to be a company. Sorry, am I allowed to say that? No, you're not allowed. No, this is a family-friendly show. It's very stupid to be a company. Like try to race to the bottom to see how little you can pay your employees. Bro, if I'm making hella money, we're all making hella money. Like, like it's I'm trying to pay them more to see if man, like maybe tomorrow we'll start being like cash flow negative.

2:00:00But I still make money, bro. Like I would like to pay these guys what they're worth. And the output is fucking insane. We did 10 million UGC views in what, like eight days. Like, like you don't see this sort of traction in any company and you don't see killers like this in any company unless you're paying these motherfuckers like what they're worth, bro. Like, I don't know why. Like, like,$130. Maxed out contracts. Maxed out contracts. Exactly. Exactly. Yeah. What about devices? I mean, it seemed like this would be a natural fit for some sort of AI wearable or other platform. Is there an app coming or are you interested in what's happening with Johnny Ive and OpenAI?

2:00:34What's your take on the device world? We're very interested in the hardware space. We've got like a million things cooking on hardware. We got people in the garage right now working on, you don't even know about, bro. like we're bringing manufacturing back to America and it all starts at the Cluey Garage. Cluey Garage. Let's go. I'd love to see it. Nobody, you know, they doubted, but you guys are re-industrializing America. You guys really are the... There's a hard time down there. They're working on brain chips down there. Brain chips. Brain chips. That's the future. There we go. There we go.

2:01:05The new Neuralink. Yeah, I mean, you know, there's a world in the future where you guys actually just roll up Neuralink and OpenAI. For sure. Bring them under their Cluey umbrella. Yep, definitely. It's possible. I'm excited to offer acquisitions for both of those companies. Yeah. It's in the roadmap. It's on the roadmap. All right. This has been a lot of fun. I'm excited for you guys. It is. And I have no doubt that you'll go from 10 million views a week to 100. And I'm excited to see you guys hit that billion view mark very soon. So keep it up. We are all very entertained and rooting for you.

2:01:41She, she, she, she. She, she. I love the energy. Thanks, man. We appreciate you joining. Later, guys. We'll talk to you soon. Keep having fun. Bye.

2:02:16You know what he needs? He is throwing the AI Engineer World's Fair, and he's going to be joining with some other folks, breaking down what's going there. We can kind of revisit this Sarah Guo post, which we talked about later, possible topics for her keynote. Apparently, she completely revised everything, but I really liked it. She wanted to talk about AI native UX, not just chat skins, vertical AI surge. That was something we were discussing earlier. multimodal frontier video 3d audio retrieval plus long-term memory synthetic data flywheels obviously super important the sim to real robotics push closing the autonomy gap robust agentic workflows cell scale digital twins and programmable biology compute geopolitics world models for zero shot planning and rl environments that actually generous generalized she has mapped out the like the true surface area, but more recently she authored a post, an article on X, directly on X.

2:03:14You can read the full thing at her X page, Sarah Normas, on Taste. And I thought this was an interesting post. Lulu quote tweeted it, and she says, Stripe returns errors in plain English. Quote, that card number doesn't look right, not error underscore invalid parameter, because developers debug at 2 a.m. Spotify's shuffle isn't random. It avoids playing the same artist twice in five songs. True random feels broken. Engineered random feels right. Notion's drag handle appears only on hover. Six dots arranged in two columns, not three lines, not always visible. Because permanence is clutter and six dots whisper, grab me, while three lines shout, I'm a menu.

2:03:56This is taste, the relentless, almost painful ability to know what should exist, what shouldn't, and where quality matters. It's the difference between shipping a product and shipping a point of view. The best founders understand that taste is a competitive advantage that compounds. It runs deeper than pixels. It's in your code base, your culture, your cap table. And I'm just thinking about the Clue Lead Taste, which is just like the most maximal possible. Just maxing everything. He's max maxing. He's max maxing. He's such a character. I love that we can have him on and then have someone, you know, like you know public company CEO on the same we got range we got range it's fun um yeah that was absolutely wild he um uh we should probably text the families and say you know don't play this in front of the kids yeah but uh he knows you know the level you know he he even in every word he's thinking of how do I get the max amount of attention out of each incremental word totally it's not just each incremental post he knows what he's doing even I I am like systemically offending people in multiple ways in a single sentence but at the same time like he's entitled to be his own personality yeah and and i do think you know not to get completely sidetracked now that we're on to the next topic but uh i do think you know an enterprising young journalist might want to take a crack at roy just because it will turn into a feud yep it will probably be good for both of them it will be very i would recommend it's wwe it's wwe paywall the article he's become a little bit of a heel of tech like yeah he's somebody that people like love to hate no it's like how do you take the yc playbook and just run the opposite of it yeah like it is it is the inverse of the yc playbook which is you know be loud before you're confident in your business model and confident in your product um the question is just like at some point getting yeah it can't all be style.

2:05:53There needs to be some substance. You have to actually build a product that people will be, that people will love. You have to make something people want. And so we're not seeing that side right now, obviously, at least not in public interviews. But I hope that he can turn it off incrementally and eventually probably need to turn it off 90 % of the time. Because yes, attention is really, really important right now. But at a certain point, you have to just deliver on the core metrics and the value prop. Or else someone else will come along and offer something with, yeah, they won't have all the style, but they will have the same sense.

2:06:28Yeah, and I think there's a venture capital firm out there that even now would give him an income, just seeing even that interview, see that and be like, I'll give this team another 10 mil. Yeah, to go figure it out. To go keep figuring it out, right? I mean, attention is incredibly valuable. But he's also being smart in that he basically has the world almost convinced that he's going to burn through every dollar he has in the next two months. Oh, totally. Which is pretty, but behind the scenes, he's like, well, I'm actually making money. Yeah, I mean, Avi Shipman at Friend was going through a similar thing where there was that big story about, oh, he raised like a$2 million seed round and he bought a$1 million domain.

2:07:09And everyone was like, oh, he's so wasteful. But then, of course, he would finance the domain. So it was really just adding like an incremental like 30k of burn per month And it wasn't really like he didn't take that whole hit up front and then he was able to kind of build a business Yeah, he's making a category bet on AI companionship. Yeah, and having friend.com is a great is Will lead to increased marketing efficiency over time. Yeah, he starts to scale so Well anyways, you should go back to this completely. You know what you should do He should get a bet. He should get a watch on bezel talking about taste.

2:07:42Bezel is how you express your taste. Your bezel concierge is available now to source you any watch on the planet. Seriously, any watch. So go to getbezel.com. So Sarah goes on to say, think of it like running a restaurant. It's so funny comparing the taste thing to this. It's like the most serious post. There's the most silly, just silly hype train over there. Think of it like running a restaurant. Anyone can follow recipes, source ingredients, and serve food, but the difference between a forgettable meal and a Michelin star isn't just technique, it's the chef's palate. Their ability to know when something needs more acid, when a dish has one element too many, when to stop plating.

2:08:18Software is the same, but products aren't just feature complete, they're composed. And we see this with a lot of the best software products. The software is more like art than science in terms of the design and how you're interacting with the user. so sarah says it's easy to say hard to do everyone claims to have taste now it becomes the new product market fit a term so overused it's lost meeting founders drop we're taste driven in pitch meetings vcs nod knowingly nobody defines it most companies confuse taste with aesthetics they hire a design agency pick a nice font and call it done but real taste runs deeper in the error message, the loading sites, the loading states, the features you killed because they were merely good, not essential.

2:09:04Real taste hurts. It's saying no to features that would triple your TAM. It's spending a week perfecting an interaction that users will barely notice consciously. It's choosing the harder technical path because the UX is 10 % better. If your taste doesn't cost you something, it's not taste, it's preference. The pain compounds daily. A Fortune 500 prospect wants a demo tomorrow? Do you ship a half-baked feature or the half-assed collateral or lose the deal? The entire AI landscape reshuffles every three weeks. Do you chase every new model or trust your vision? Your competitor just shipped something flashy.

2:09:40Do you match it or hold your ground? Anyway, we'll have to have Sarah on the show to recap her talk. But right now, I believe we have Sean coming in to the studio to tell us about the AI Engineering World's Fair. We're very excited to have him join the show. Thanks so much for hopping on. I wish I could be there. I have the biggest FOMO I've ever had with any tech conference because the lineup seemed absolutely fantastic. Unfortunately, I have a large family and a studio that is covered in junk and we're moving in. So I appreciate you taking this remotely and we'll have to do the next one in person.

2:10:13But thanks so much for joining today. How are you doing? Hey guys, glad to be back. It's funny because like your setup looks so great on camera. I'm just imagining the mess that it's off camera. Oh yeah, yeah, yeah, yeah. It's definitely a work in progress. But part of the brand is of course just showing the nice parts. But congratulations. This is many years in the making. Can you give us a little bit of the history, the plan? And then I want to go into some of the hot topics that you've been discussing today. Yeah, I guess this is the fourth conference we've done. And I started this basically right after the sort of chat GPT moment and talking with enough developers and understanding that the people who can wield LLM APIs are going to be way more powerful than people who just chat with products.

2:11:02And this is actually going to widen a lot because they can basically wield serverless intelligence. And so I coined, you know i sort of popularized the term ai engineer andre karpathy sort of endorsed it he said that um he does believe that you can get very far without ever training anything which is a big thing for him to say because i think the status quo at the time was you know it's high status to trade models totally scientists but now it's actually consensus now that you want to work on wrappers rather than models and uh there are many many multi-billion dollar companies that have spoken at AI Engineer that reflect that fact.

2:11:37And yeah, so like, you know, I guess - And there was a big shift there where there was a moment where like the vibe had shifted a little bit to like the application layer, but there was still the idea that if you were going to build an AI legal startup, you were going to train an AI legal foundation model. And now it's moved all to post-training, all to RL and how you're prompting and how you're integrating and ultimately like the user interface. And so, yeah, it makes a ton of sense. And I'm sure there's a ton of companies that are beneficiaries of that boom. Yeah. And it's not even companies, it's more just like customers.

2:12:14Sure. Because the foundation model labs are never going to work directly with like your healthcare system. Like it's going to be like a bridge that comes up and does that. The foundation model labs are not going to work directly with lawyers. It's going to be Harvey. You know, I think so that that definitely led into the rise of vertical AI enabled SaaS. And I mean, that's consensus now. Yeah, I'm sorry. That's a Slack thing that by the way, the conference is still going on. We just had the morning keynotes. I have my chat with Greg Brockman later today. Fantastic. And he doesn't know this, but I can, I might as well just share this, but like we have a nice little cameo from Jensen Huang coming by.

2:12:53No way. That's great. So we're trying to level up the cat, the The keynote this morning with fantastic, like Microsoft, you know, they like really went on. They're going so hard after the AI audience. We were just reading about that in the information. Platform, platform, platform, platform, platform, says Satya. Yeah, yeah. I think they see this as their chance to overtake AWS. That's great. But like AWS is also, you know, also sponsored and like having a very strong presence with us. So like we just want to be the vendor neutral place, right? Like all the big clouds, all the big labs work with us.

2:13:26We have the MCP team here with the entire steering committee presenting as well. And we just want to be for developers. This is kind of the trade show. You come to hire people, learn about what's new, and upskill. I have a couple more questions. How long do you have exactly? Five minutes? I can go until like 1.30. Okay, great. We had a guest and I moved into a different day. Okay, first. Perfect. So I want to know about the mix of attendees. How many folks are trying to start venture-backable application layer AI companies versus is there a new trend of someone who's building more of a vibe-coded, almost lifestyle-type AI-driven business?

2:14:12Are there folks from either companies that have established themselves and are now trying to bolt on AI or are there lots of folks that are working for large companies and just want to stay ahead and become AI engineers? What's kind of the shape of the audience, if you can characterize it at all? Yeah. So we do surveys, but we don't know specifics to that high level granularity. I would say about 50 % are people working at medium to large size companies and trying to upskilled and then the others are uh smaller companies uh and you know our most popular title is like founder or cxo of like a a smaller startup and those are those are venture backable and i think mostly that's just a function of us being in san francisco yeah because obviously we we will have that startup bias um i do think that um one thing i kind of don't really care about this whole lifestyle versus venture back thing because uh for example i have a tiny teams track that's speaking this year.

2:15:13Tiny teams is something that I'm trying to push as an idea of companies that have more millions in ARR than employees, right? So your revenue efficiency is so high because obviously if you pay each employee less than a million dollars, you're probably profitable and therefore you don't actually need the venture money except to point to marketing. And that's your choice. You can be profitable. I have a six-person team making more than$40 million. And yeah, I mean, And it's absurd. The amount of leverage you can get with agents and also building AI products for other people to use where you're just kind of passing through or slapping a margin on top of the tokens that you resell from the big model labs.

2:15:54I think that really makes a lot of sense. Totally. Has the narrative of like, oh, if you're building an AI, like you're going to become obsolete by Google or OpenAI is going to steamroll you in their next Dev Day keynote. Has that narrative dissipated? And if so, what's driving it? Is it kind of the pre-training scaling wall that we're kind of seeing with GPT 4.5? Are you guys still – I don't know if people are still talking about that. I mean, everyone – Everyone's moved on. Yeah, we've moved on to a different time, right? Like, yeah, I think most people have agreed. Like, there are still new pre-trains happening, especially with the open source model that OpenAI is working on.

2:16:35Sure. But yeah, I mean, I think we've just like, we've seen it come and go enough times. So for example, we have OpenAI launching Chatapakai Codex, which is just head on a direct Devon competitor. You know, Devon's not worried because they have been doing this for two years and they're much more polished in terms of the integrations and they have different things that their customers already like. And so I think it's just like everyone is going to need their version of a thing. And so this is the sort of house store brand version of what ChatGPZ codecs could be. And it's not competitive just because the ocean is so huge for software engineering.

2:17:12And Devin has at least established the category by being first there. And I think you can see similar versions of that across the domain. Yeah, I haven't really seen anything there yet. Although it's not to say that it doesn't happen. It does happen, for example, with the first wave with GPT-3 startups like Jasper. but it I mean yeah so far there's no fear that in fact that people are very excited to meet the foundation model labs like this is where the engineers meet the lab people and lab people train them on how to use their LLMs and I think that's that's a perfectly harmonious relationship to be honest I haven't seen it what was your reaction to the the news around Anthropic and Windsurf yesterday a little bit of a bit of drama on the timeline but yeah yeah give us uh give us a walk through for those who might not be familiar of what happened and then I'd love your analysis.

2:18:02Yeah. So the history of this is that Windsurf is an independent company that basically kind of followed Cursor's footsteps and launched an AI agentic IT. They've done very, very well for themselves in a very short amount of time. I think they launched, we're the first podcast that they launched with in October or November last year. And then there's rumors that they got acquired by OpenAI for$3 billion. Those are rumors that are unconfirmed by both sides. I've talked to both sides. Yeah, I've been fascinated how everybody just takes it as fact. I take it as fact. He wore two polos. Yeah, he wore two polos.

2:18:38But again, at no point, every news outlet, legacy news outlets have been reporting it as fact, even though no side has verified it and nobody said it closed or anything along those lines. Yeah. The only thing I can say is Winsurf is speaking tonight right before Greg Brockman goes on. And there's a reason. Exciting. We're not dropping a ton of alpha, but I'm just saying that's all I can say. That's all I'm allowed to say. Yes, of course. And I think that so there's a relationship there, as there is a fact that both Cursor and Windsurf have benefited a lot from their relationship with Anthropic because 3.5 and 3.7 have for whatever it's worth been regarded as the best coding models.

2:19:26Yep. OpenAO will disagree. Gemini will disagree, but whatever. Like the community has voted. Yeah. So overnight, I think like two days ago, one day ago, Anthropic cut off the first party access to Claude for Claude to Windsurf. This is their top model. And now they just don't have access as a first party tool. You can, for example, still bring your API key and use your accounts on Windsurf. But it's right. You can't just like like Windsurf, I'm going to pay you 20 bucks, use your account, whatever. You know, I just don't want to worry about the rate limits and stuff. That's gone. Windsurf, they just woke up overnight and that's gone.

2:20:01So a lot of people are very upset. This is a very big no-no if you're aiming to be any sort of credible LLM API provider to just cut off access. Google hasn't done that, even though if you take the rumors of the acquisition on face value. But I think this does lead credence that I think Anthropic at least thinks that Windsurf is just a competitor product now. So I think insofar as you're putting odds on whether the acquisition actually closes, the odds have moved up. interesting yeah the other the other factor that's interesting is how much does this benefit cursor if you're a developer who loves various clod models and now maybe you have a reason to go spin up you know market share of cursor versus windsurf have you seen like estimation estimations what is that i i think the numbers i've seen again i really i saw it on my timeline on twitter i didn't save it and just go fast but someone please look this up it's something like windsurf is five percent of cursor wow because based on the based on the market caps i would assume that windsurf's like 30 i know of course so the thing that you don't see there is cursor has entirely developed business on the ide sure where windsurf used to be codium which used to be like this gpu service gpu company that has significant lm inference uh for enterprises like they they spent the last four years doing that yeah yeah um so they're buying that team that revenue that product as well as Windsurf.

2:21:26Oh, interesting. Yeah, that makes a lot of sense. How do you think that the agentic coding market is shaping up? We've been kind of talking about maybe it's three different markets. Like I can go to GPT, like I can go to O3 and I can ask it for a question and it will just write code and write Python and execute it. And I don't even have to tell it to write code. It's just if I ask a question that requires code, it will write some code. Then of course, OpenAI now has Codex, and then they also potentially have WindSurf, or you can think about the IDEs as a different entry point into that market.

2:22:03Are we seeing a true permanent bifurcation between synchronous and asynchronous AI coding usage, or do you think these all blend together at some point? Yeah, I think that actually you're just catching up on something that has already been a thing. that was the divergence and actually the converging. So it already started that you had the synchronous ones like Cursor Windsurf and then you had asynchronous ones like DevIn and Factually AI, for example. And they were all sort of like what we call the developer inner loop, which is hands-on keyboard coding, and developer outer loop, which is PR review and all that kind of stuff, or like file an issue, make a PR, that kind of stuff.

2:22:46So those are merging. The explicit goal of Codex, it's something I'm going to talk to Greg about later today, is that they will merge those paradigms just as they're merging the reasoning models and the sort of instant thinking models with GPT-5. And that totally makes sense. It's technically really hard to do because one's on your machine and one's cloud. And also they're just different set of user experience paradigms. Is my video freezing, by the way? It is. You look great. You look great. I see you guys. How seriously do you take Google's coding agent, Jules? Do you think it's just an experiment that they're putting out there?

2:23:24Do you think they'll actually invest in it and try to get real adoption? You're trying to get me in trouble.

2:23:32You're conflicted. You're conflicted. I'm sure Google is a part of it. Google's great. products wise they have had hits like lm that then have failed to continue the momentum for whatever that sure that i and i think that is i think that most people are extremely unfair to notebook lm i think they've shit really well it's just that you will never repeat the initial wave of excitement that they had totally just never will like yeah and i mean notebook lm seems interesting because that feels like a case study in if you're a startup and you're fast following an innovation that actually came out of a hyperscaler, that's where you turn into a bullet point on the next dev day or something like that.

2:24:16Because Notebook LM, it felt like there was this amazing, tremendous momentum online. Everyone was excited. And then it was kind of faltering and it was like, oh, there's no app. So maybe I'll build the app. And I'm sure some people built Notebook LM apps or Notebook LM spinoffs. And then it just took six months. months and then there now there's an app and now and i'm sure they will reiterate on this and so yeah yeah i i know to some degree uh but but but i i think the lesson is probably like yeah there's probably some narrow window where you can cash grab as kind of like a cynical like ripoff startup but like realistically if you if you found out about the concept of turning uh a deep research report into an audio product from notebook lm like you might be just behind the ball and so You don't have the right to own that market long term.

2:25:02As opposed to, I think, Devon, Cognition, they have more of a right, and Factory, they have more right to continue to fight with Google's Jules product. Because they were really there earlier. They issued the original hype and brokered a whole bunch of enterprise contracts and kind of got a couple years down the road before. So they're not like a fast follower to big tech. they're actually, big tech is maybe fast following that. Yes. I think it just boils down to execution and everything, right? Yeah, of course. And I think that's one of the reasons why we've actually somewhat broadened out from just engineering to we've added AI product management and AI design.

2:25:44Okay. And I think that really good, this is product management. This is just straight up, can you keep up the momentum? Can you listen to your users? Can you come up with creative new stuff that keeps the momentum going? and some teams can and some teams cannot. The only thing I can say with regards to Google and the fate of jewels is the initial, the founding PM of Nobook LM left and she's speaking here with her own startup. Oh, yeah.

2:26:11It's really hard to keep employees, people who start interesting products for Epic Lab because they will get any amount of money thrown at them. Yeah. Veg Capital, let's strike again. $20 million seed round,$10 million of secondary if you quit your job right now. Probably not that extreme. I wanted to ask you about GitHub.

2:26:36How much do you think Satya cares about GitHub? Do you think he's really pushing the team on that? Yeah, is that an important wedge into the AI coding market? Because obviously they were first with GitHub Copilot. And you can imagine that it's just a phenomenal distribution channel. And when you look at the value of, and the way Microsoft executed around Teams rolling into, you're already using Outlook, now you're using Teams instead of Slack. You could imagine really great adoption, maybe not at the most cutting edge companies that are hyper online, but we could see one of those charts where we're like, oh wow, Microsoft's really crushing it in developer tooling.

2:27:18I mean, absolutely. I think it's extremely core. What would you put GitHub Copilot revenue at right now? If you had to hazard a guess. 500. 500 million? Yeah. You nailed it. Let's go. Standalone, that is a publicly listed company. Wow, yeah, yeah, yeah. That's like the perfect bar for not building it. Here's the challenge, right? Here's the challenge is that no one is talking about GitHub Copilot. Is that a challenge? It's inside of Microsoft. I mean, no one's talking about it. No, I know, but nobody's saying, look what... Nobody's sitting there being like, I'm blown away by Copilot. I think that people who live in the Microsoft stack are, because that is what they have access to and their company agreements are.

2:28:10I think on Twitter, there's a novelty bias. People always want the new thing. They want to support the little guy. I support the current thing. You want to support the current thing. But okay, here's the mental framework I want to leave you guys with on that stuff. Data gravity is a thing. Data attracts the compute, attracts the money, attracts stickiness. Once my data is one place, everything else moves towards that data. And if my code is the most valuable form of data I have, that's the most expensive data to acquire. So if my code already lives with you, then it's much easier for my coding agents to also live with you.

2:28:52And so they have a home turf advantage. I think the GitHub acquisition back in the day was one of the smartest and most enterprise acquisitions in developer tools history. So I think you should take that point of view. So that moat is so strong. Try to be like a no-name random dev tool startup that comes up and says, hey, give me access to your code base. I need to run agents on it. they're like no like i i'm already yeah what happened what happened with codex yesterday there was one user that was reporting that uh private repo information was leaking from one to another was that user error did you did you track that at all now you're probably getting uh no i i was i was running the conference uh yeah and individual things are going to happen yeah of course happen um i think github's reputation is going to uh last it through that I think it has to be really egregious for that kind of stuff to stick through.

2:29:46I mean, I think like more broadly, I think that there's a standard stack of what we are calling, what Andre Karpathy is calling the LLMOS, right? That I think you guys are kind of going and learning about. But I think it's pretty well established by anyone who works on software agents, coding agents, which is you want a sandbox. You want a browsing environment. You want branching. You want fine tuning on your code base. like there's just a standard stack of things that cloud code codex jewels cognition um um factory and anyone else in this field they're all converging on the same thing and it's just who does it best and who reaches who solves it for their customers best i think that's that's going to be the name of the game for now makes a ton of sense uh do we have to roll over to the next person uh no because Okay, you're good.

2:30:34Yeah, I mean, I guess my question on data gravity is, it seems like if we compare the evolution of video generation models, Google has a really amazing cornered resource in YouTube because that data is so big. It's so many tokens. It's growing. It's growing, and it also hasn't been like exfiltrated to the public web. But it feels like GitHub, although it is a massive data set, it's not nearly as big. I think somebody clocked it at like 100 million tokens or something. It's just not. No. No? How big is it? Way bigger? Well, public repos, because I don't think that they can train on private repos, right?

2:31:17Or can they? Who? GitHub? GitHub? GitHub? GitHub. I don't think they will let themselves. Yeah, they wouldn't. Yeah. I don't know. 100 million just sounds way too low. That's my gut reaction. Sure, sure, sure. Keep going. So the question is like, how durable is the data moat at encoding versus in video generation? Purely, I'm just thinking about the raw data set size of YouTube has to be orders of magnitude larger than GitHub. And so I would imagine that Microsoft probably has a less durable advantage versus Google's like kind of march down the video generation pathway. way if it's truly restricted by data.

2:32:01Not all data is comparable like that. So you're comparing verifiable data to unverifiable data. So code can compile. And if it runs, you can do more code like that. And that creates the RL loop that lets you generate synthetic data for more code. With videos, you have what you have. And you can train on that. And you're technically not even supposed to train on that. But who knows what Google is doing behind the scenes. Well, I think Google can train on public YouTube videos, right? I don't know. I imagine that like I've posted videos on YouTube. I imagine that I've opted in at some point. I think there was like a big fuss with MKBHD and all the other guys.

2:32:39Yeah. Taking up a fuss about this, you remember? Sure. So I don't know. I'm not a lawyer, but like I'm sure it's a completely untested clause that just has to go to court. I just want to say to Google, if you remove John's rate limits on VO3, you can train on our back catalog. Full permission. Please. Yeah, I think there is that diversity of opinion, right? Do you want to just give yourself to the machine or do you want to keep your data to yourself? And it's like there's no in between. You're either one or the other. It's a very strange dichotomy. Many things in life are a spectrum. This one is not.

2:33:15Like you run into someone, they're either privacy maxi or they're like an AI maxi. Yeah, yeah, yeah. No, that makes sense. What else is on the cutting edge of debates that are kind of raging at the conference this year versus prior years? We've lived through like the P-Doom debate. We lived through like the Leopold Aschenbrenner era. We've shifted to the geopolitics debates. But what's kind of on the frontier of like hottest topics to discuss? First of all, Leopold was right. And I don't know if people know that he was on Ilya's team when the sort of board drama happened and he was like directly connected to it.

2:33:55Anyway, we haven't lived through it. We are living through it. It is happening. It is directly leading to the geopolitics because he was foreseeing that and he was exactly right. Yeah, but the way I would kind of characterize at least my takeaway was that his piece was a was a little bit of a pivot from the paper clipping doom and a shift to geopolitical competition in AI. Is that a mischaracterization? No, nailed it. But he was right. Yeah, no, no. I completely agree. I completely agree. I'm just wondering if the consensus is that he was right, then the book is closed. We're not debating that anymore.

2:34:33What are we debating? What's more modern? What are the debates? How to do great AI PMing, how to run a tiny team, We have a robotics track for the first time that is Tesla Optimus is speaking, physical intelligence, Waymo. Waymo just overtook Lyft. Yeah, I saw that. I saw that already. Voice is the hottest thing in terms of modalities. Everyone's sort of building with voice because I think it's finally good enough. And I think maybe the last thing I'll highlight to you is we are also emphasizing security for the first time. Security is kind of a boring topic. Nobody really wants to talk about how to secure your system.

2:35:10but like they actually do now because they have real money running through their their product so there's all that and then that is like roughly important equal in size to the excitement about mcp and so we have an entire mcp track with the anthropic team here very cool because it's uh because they're nice enough to to come by and that fills up the the whole ballroom that we have so are we gonna get payments in mcp is there a sub track for putting stable coins in there or something like that? There's nothing in the official spec, but we have a number of people and we have a speaker talking about the sort of MCP economy that's being enabled.

2:35:45I do think that remote MCPs and authorization give you the foundation for basically just remote agents that you can buy and hire. And that's effectively what it is. It's just that we're still in that stage where these things are still trivial enough that you can actually just write your own. So you have to overcome that build versus buy for this to actually kick off. Okay. And I would just say, like, I'm not at all a crypto person, but I would say that the crypto people have been ahead of us here and they haven't found that much yet. You know, no, but they're actively hunting. And if anyone finds it first, it will be them.

2:36:20It's circles presenting here. Solana presented in my previous conference. And they're all they're all on it. Stripe is also on it with their stable coin thing. Something's going to happen here. Yeah. In robotics, are we getting to a point where we're starting to see an ecosystem of companies pop up like we've seen? No, it's just. Yeah, so they're all vertically integrated. Sure. They're all building their stack. I don't see like any horizontalness. Nothing. That's what you're going for. That's exactly what I was asking. Yeah. Yeah, it's so weird. I think, you know, there's just so much custom needs that you have to sort of reinvent the universe every time.

2:36:58The one thing that does reuse bots is CloudChef, which is a very young company, but I ran across them when I tried their food. So they're a kitchen robot. They do demonstration learning from a single shot from, let's say, like a Michelin chef. And you just give it the ingredients and it will just flawlessly just kind of repeat that cooking for you. And you can hire it for$12 an hour. So it's meant to directly replace human labor, which is very expensive and very labor intensive and unreliable. Let's put it that way. Sure. And they use they can they can live on top of any other robot arm or Tesla Optimus substrate.

2:37:37Got it. OK. Because it's more about the robotics sort of framework than it is about the individual hardware. And I think like that is the first time I've seen that happen. I'm like relatively new to this, but like this is the first time I'm like, oh, OK. Like he's actually pretty confident he transfers across any system as long as they have the minimum required set of device drivers, basically. And I'm like, yeah, that's cool. Like, yeah, yeah. So he's speaking tomorrow. What about other pieces of the robotics data stack? Are we thinking about like data brokerage or kind of like a scale AI or maybe scale AI actually working to generate more robotics data?

2:38:15Anything on like the sim to real gap? I saw some semi-analysis summary of a paper that was pretty cool. so i would say that as an as a sort of industry practical conference those are just in the domain of research right now it's in research yeah yeah it's like there's there's a lot of papers out there um jim fan had a fantastic talk at sequoia sense that i recommend everyone watch if you haven't seen it about the physical turing test yeah how you need to uh just do a lot more simulation and there's a there's a lot of work on being done on this um i think google genie is the other one that i would recommend people to there's like a really interesting tie-in between like the generative video world and the robotics world that you wouldn't necessarily expect until you like spend some time in that um but we just haven't focused on it because people cannot get jobs in it like i i want people to get jobs at my conference that's like almost the whole reason that's the whole point i bring the companies i bring the engineers they meet they they fall in love with each other that's like to be that's like the most fulfilling thing can you give a a high level update on the hiring market today?

2:39:15What's fact? What's fact? What's fiction? It's one of those things people love to talk about how bad things are. And yet every company that I know can't find people. And maybe there's a talent kind of gap there. But what's your read? Yeah, I mean, I think both are right, which is weird. This is actually the topic of my conversation with Greg Brockman later today. because half the people I meet at my own conference are worried for their own jobs, right? They do ask about this. It's not like they're blind to it, but they're just like, I'm here to see what else I need to do or how does my skillset need to change for the future?

2:39:56And nobody knows. Everyone says the trite thing, which is like, oh, I'm going from, I write the code to I manage things that write the code. So I become an engineering manager. But like, what does that mean in practice? how much knowledge do you need to supervise an agent? You actually kind of need a lot when it goes off track. So that's super unknown. I do think that the juniors, the fresh out of college students, they're a little bit cooked.

2:40:26Unless they're good. So it really amplifies skill issue. It really, really amplifies skill issue. Yeah. It's a hard thing to try to tell somebody you just need to be better five times smarter than you are and then the job and then you'll have like a bunch of offers but i also think i also think that the the solution here and and we we just hired uh an intern who already shipped a new product he he started uh monday he just shipped uh something relatively simple but it's cool it's a guest directory he shipped he just sent us a v1 of that product that he built last thursday we saw it and we were like this is super cool why don't you start on monday and it's that one kind of interaction that can get your foot in the door that allows you to develop skills and build relationships that you know hopefully we you know we work with him for a long time but even you know i'm sure people are already seeing him at tbpn and we'll probably try to poach him uh before the end of the summer so uh it is it is like a changing shape of the software engineering employment because like i I feel like the vibe coders can come into organizations that might not have a full engineering department and then have an impact because of the way the tools work.

2:41:43But it's certainly a lot of change very quickly. I would say that's not the consensus yet that hire a chief vibe code officer because humans are still very much needed to patch the gaps that the models are not good at. And it's really painful when they do go wrong. And they do go wrong. I think all the hypes that you see on Twitter fail to admit what happens one month, two months, three months after. And that's honestly what you get paid to do as a software engineer, make maintainable software and not refilled all the time. I've even seen people repurpose old projects and say, I vibe coded this in 24 hours.

2:42:21And everyone's like, wait, now this was published in 2019. You spent months on this and you're just hyping this up. That's misinformation. I can't govern. I pay attention to what people say there, but I do care what people report for themselves that they've done at serious companies. So one thing I would highlight there is Booking.com did about 10 company years worth of migrations in three months with their sort of automated code migration that they did with Sourcegraph. So we did a talk with them at my conference in New York in February, which if you want to look that up. so like those that company has been around for like 14 15 years that's that's a serious code base right yeah totally they're just reporting their success they got nothing to gain from selling you on it they're just they're just happy about it and they want to teach others yeah so you want to look for those they're not trying to sell you anything they're just reporting like their progress at a real company yeah and it's that's really what i try to optimize for for the for the conference it's it's hard because not everyone's incentivized to do it you just have to create an environment where they get something out of it by meeting their peers at a thing like this.

2:43:32Well, give us the plug. Where can people watch your talk with Greg Brockman later today? Yeah, it's on YouTube. YouTube at ai.engineer.dot. And yeah, what would you ask him? I mean, you're having him on eventually. So what's the way that you would open the conversation? Ooh, that's a good question. For Greg, there's a lot. But I mean, the question that I feel like is at the top of mind in like Ben Thompson world this week is the shift from training to inference and how workloads are shifting. This was in the context of an NVIDIA earnings, but Greg obviously has intimate knowledge. And so OpenAI is obviously going to do larger and larger training runs, but they're also doing tons and tons of inference.

2:44:31How is that shifting? And then what I want to know, and I don't know how much he can speak to this, is like where, how will, if we get to a world where we're in 80 % inference, 99 % inference, something really extreme, does that change the type of data centers we're building? Does that change the chips that we're demanding? Are we moving to ASICs? are we moving to FPGAs, something like that to speed up the actual inference workloads? At what point do we actually bake these models down if they're deemed good enough and then we're orchestrating them? If we hit a plateau, that might make sense. At the same time, if there's a lot of promise on algorithmic progress and we're expecting like, oh yeah, we're going to leave the transformer behind eventually.

2:45:13Well then, yeah, you don't want to bake all that down. So I think that that's like a little bit more of like a semis question, but it is an interesting question for him as well as he's like seeing the workloads. It is very relevant. He does have a role to play in Stargate that I'm not super clear about. I'm going to ask him about. Yeah. And yeah, I mean, I would say that nobody's betting on the endotransformers, except for a very small number of people that are just experimental. And it's really about just scaling the RL. Yeah. I mean, there is this interesting thing that's happening. I believe Images and ChatGPT seems to be using a number of different algorithms combined.

2:45:51So there's a little bit of diffusion in there, potentially. There's some transformer-based stuff. I don't know if you're pushing back on that. I'm not. There's a strong hypothesis based on the hints that they have dropped from the people that worked on it. Yeah, that there's multiple algorithms at work, right? It's a diffusion head with a transformer backbone. The rest of it is just regular 4.0. And that's why I'm so good at chat and has decent world knowledge. Yeah. There's not like a ton of complexity there beyond that, that we'd know of. Yeah. And so what I'm interested in is like, we saw the demo from Google on diffusion, text-based diffusion, 900 tokens a second.

2:46:29I've heard good reviews about that. I've heard kind of mixed reviews on that. Maybe it's not a path, but what does the future of an LLM look like if we're applying the same path that we've seen in images, where we're seeing an ensemble of models come together to create a better, like even more multimodal, multi-architectures, because it feels like we're moving further and further away from the single big transformer being the answer to everything. And so what does that mean? Are we going closer or farther away from the single big transformer that is like the god in the weights. Yeah, this is where I can offer a little bit of coaching for your audience.

2:47:13And maybe you guys have talked to a lot of people, so I don't know. This is where people usually say the term, make sure of experts, and they're wrong to use that. And sometimes they say, make sure of agents, make sure of experts. They're incorrect. Anyway, we have talked with a lot of the Frontier Lab researchers. none of them believe that that is the way forward. That is an optimization for the current thing. Which one? A mixture of experts? Like a mixture of architectures. A mixture of architectures. Okay. Yeah, so experiments, for example, with Jamba from 821 in Israel where they mix, for example, like a Mamba layer with Transformers.

2:47:56Sure. That seems to be promising, but even Gnome Shazir, a character was just scaling out Transformers I asked a variant of the question you just stated to Gnome Brown in an upcoming podcast that we did, we recorded it, we haven't released it he's the same way, he's like anything you're trying to do fancy around mixing up weird architectures is just not going to scale, just scale the basic thing and he's pretty strongly convicted in that I know people I can't name at Gemini who is also pretty strongly convicted in that I have a reason to suspect otherwise Obviously, you can't shut down anything because it might be true.

2:48:33I'm just telling you what the people working on this would say. Because that's the same question. The last thing I wanted to kind of cover, I would want to get a sense of prediction markets are pricing in that Google continues to dominate on benchmarks. How much is OpenAI even going to... How much do they even care about being at the top of benchmarks? Is it purely an ego thing or is it just about, you know, usability and value for users and code quality and things like that? It's usability and value. I think you need to notice that the benchmarks that Google talks about at Google I.O. are no longer the benchmarks that OpenAI measures itself on.

2:49:19And, you know, I want to stay friends with them. But, like, there are some that are more sus than others. Sure. And I think at the end of the day, the customer will win. The only one that has clearly been caught out kind of lying, kind of like being whatever, underperforming, I got to say it, it's Apple. When they launched Apple Intelligence, we were all very excited about the Apple Intelligence paper, which is beautiful documents that had no evals apart from their own internal evals. So they were not accountable to the standard set of evals that everyone else has. step one is you should try to hold yourself to some public benchmark just do it so Apple did not do that and then now there's a question of are you holding yourself to a benchmark that can be gamed and there's been accusations of Ella Marina being gamed and unfortunately that's what Google has put all its chips on they may not do that for Gemini 3 but for 2.5 that's what they did and probably because they spent the last year doing it It's all revenue from here on out.

2:50:24Like that is the benchmark. Revenue is the benchmark. GitHub Copilot, 500 million. Size gone for GitHub. Revenue is also tricky. Yeah, you're right. You guys know how, like Gemini, I don't know if you know, Gemini gives a billion tokens a day per human. What? For free. Your personal API token. Okay. I'm shooting you. You can set up a camera right now, run Gemini 2.5 flash on it, run like a frame a second and ask you to do whatever you want for free. Wow. It's absurd. So they are market cheap. Thank you. Intelligence too cheap to meter. We love to hear it. Okay. Like that chart that you saw at Google I.O.

2:51:02where their tokens went like this. Now you know why. It's free. Yeah, yeah, yeah. I mean, they also have a very, very good model. Don't get me wrong. But I'm just like, revenue is a short-term play. Like if you're trying to maximize revenue now, you might cut off yourself from getting the largest data set in the world, which is the early adopter humans that are going to give you their data because they're like, train on me, daddy. And you're giving to me for free, like whatever, train on me. And that's a deal that lots of people, I think it's a theme that we talked about, it's lots of people are going to make.

2:51:35It's free. Why? They can eat your data. Codex is free. Why? So I talked about this in one of my recent posts like the the marginal cost of software has is has gone down to zero and like can it go negative will i start paying you for you for me for you to use me absolutely because then you'd become my labeler yeah you're you're a labeler i can never pay for you know usually have to like hire someone like the philippines or something we've maxed them out so now i got i gotta i gotta pay someone in silicon valley to label my my software so i might as well just give you my coding agent for free or like even honestly just pay you for good feedback on my coding agent.

2:52:16And why should that, why should it stop at like$20 a month when like there are people in New York who I've heard, by the way, get, they're like, they hire former bankers, they hire former hedge fund guys paying them 500K a year to label. It's wild. Well, thank you so much for hopping on. Good luck with the rest of the conference. Let's make this a regular thing. This is fantastic conversation. Thank you for taking the time from your event. And we'll be there next year, whether you like it or not. Fantastic. You're invited, for sure. We'll talk to you soon. Thanks so much for having me. Thanks, Sean.

2:52:47Bye. Cheers. Fantastic. Well, we got to tell you about Wander. Find your happy place. Find your happy place. Book a Wander with inspiring views, hotel-grade amenities, dreamy beds, top-tier cleaning, and 24-7 concierge service. It's a vacation home, but better folks. I want the TBPN army to clear out every single wanderer all of summer. Market clearing order in now. This is an interesting one. Oscar has joined the Fortune 500 for the first time. I know what you're thinking, and I agree. What, only now? And Josh Koestner says, I'm deeply proud of the tenacity and persistence of the Oscar team. We've been humbled many times from down 95 % from our IPO to now entering the Fortune 500.

2:53:33I am excited for all that is ahead of us. Wow, congratulations to Josh Kushner. That is a phenomenal run. Amazing. And somebody else was quote tweeting this saying that there's only two people who have incubated a Fortune 500 company who are actively investing. And it's Palantir with Peter Thiel and Founders Fund and Josh Kushner with Thrive and Oscar. So what a phenomenal run. It is rare that the incubations are starting to work and the incubations will continue until morale improves. Until the global economy is growing at 10 % a year. Anyway, every person who works in tech needs at least five non-tech friends to interact with closely on a weekly basis so they can understand how the general public actually thinks, says Kit Volta.

2:54:19You posted this from a different account. It was a copypasta. Somebody copied the whole thing. So I found the original. You shared the fake one, the ripoff. I shared a ripoff. Yeah, you shared a ripoff. and it got community noted and I found the original one read it here. Wait, so somebody just stole it? Yeah. Yeah. They didn't even screenshot it and put it in Bangor. That is just wrong. But 20K on this post, the repost has 15K likes and so it got, but it got community noted. This is a copy paste. You feel like that could be, that could be like automated in X where it's like, this is the exact same text as has been posted elsewhere.

2:54:53You should just know that you should maybe go to the source on this if somebody's posting the exact same thing. Except, you know, Josh Dynan will get that every day because he's constantly boasting, good morning, we are going to win. And so he would have the note every single day. JT Jira Tickets says, web design? You mean digital physiognomy? Very funny. Oh, you put another one in here from JT. Guy who is burnt out from six figures, five hours a week, big tech job. You got to just hit the roof and just get some sun. Anyway, Pavel has a good one. I don't think Waymo can work in New York City because during rush hour, you need to break the law to get anywhere.

2:55:29And I'm not sure how you could get away with embedding that in software. And IB says, Travis would have gotten it done. Never too late. Somebody says full self-driving gladly speeds for me. I thought this was good from Will. Two thick scoops. Okay, sounds good. Thank you. And it's Jersey Mike's order pickup at 1248. Dutch government collapses over migration dispute. Why is this? This is a Gmail. Oh, it's just rolling up multiple emails or something. This is Apple AI summary. Oh, yeah. Apple AI strikes again. Apple intelligence is, the summaries are so good. It's genius. 75 ,000 likes, that's a lot of attention on that.

2:56:08You got to upgrade it, folks. You're missing out on entertainment. I would pay Apple to help me produce Apple intelligent viral bangers. Yeah. I want to wrap up, but we got to say congratulations to Jesse Michaels. He got on Joe Rogan. He said it was the honor of a lifetime to sit down with a goat himself, Joe Rogan. Many people have called him a young Rogan. I've called him that. He is fantastic. He's been on a fantastic run with American Alchemy. His podcast, he's climbing the charts. If you're into aliens, AI apocalypses, Bob Lazar, nuclear weapons, secret anti-gravity research, go check out the Jesse Michaels Joe Rogan Experience episode.

2:56:46It's episode 2 ,331. Rogan's putting up big numbers. He's giving us a run, but we're catching up. We're catching up. It's wild. Just a couple years. Lastly, I just wanted to cover this quickly. AMD has acquired Breem. Oh, okay. It's been reported today to help reduce NVIDIA's market dominance when it comes to AI hardware. Talking to George Hotz, and we're going to have him on the show, but he's been back in AMD. We got Dylan Patel coming on the show Friday. We'll ask him more about AMD and their plan to unseat CUDA as the dominant AI platform. Yep. And we have other news. We are officially, we have doubled the size of the show.

2:57:27Once again, Jordy hit that gong, hit the real, hit the real gong for me. Grab this. We're going to gong cam. We got 64 ,000 followers on X. Thank you for everyone who's been around. Oh, there we go. That's great. Thank you for watching. Thank you for supporting us. It's been a fantastic journey. It is a pleasure doing business. We got the whole crew in the studio. We got a whole bunch more cameras here. We're growing every day. Damn, we got the studio cam too. The studio cam's going. The boys are cooking. Thank you for watching. Thanks for all the support. Leave us a five-star review. Apple Podcasts and Spotify.

2:58:01Especially on Apple Podcasts. Yeah, I feel like we're doing low there. I guess we make such a video-heavy podcast. Yeah, it doesn't really make sense. But if you're on Apple and you're listening and you haven't left us a review, we'd love a review. So thank you so much. And we'll see you tomorrow. We have an awesome day. We have a bit of an AI day coming together with folks from OpenAI and Anthropic coming on. It should be a great show. So stay tuned. We'll see you tomorrow. and have a great rest of your Wednesday. Goodbye. Cheers.