SPEAKER_00: we are not going to make any change unless an AI company comes to the table with money. I didn't want to do this in a way that wasn't transparent to our community.

SPEAKER_01: But now that it's out, we can literally start having these conversations. And the conversations kind of don't make sense until you have a number, right? Are we saying we're going to pay every story one penny? No one's going to care, right? Is it $5? Well, some people are going to start to care. Is it the best stories that show up in AI results over and over again? Are they going to make $100, $1,000? No, that's going to catch people's attention. And so we have to start the negotiation to see where it goes.

SPEAKER_03: This Week in Startups is brought to you by DevStats. Check out DevStats today and get 20% off plus access to their dedicated Slack channel. Just go to DevStats.com slash twist. Gusto. Check out the online payroll and benefits experts with software built specifically for small businesses and startups. Try Gusto today and get three months free at Gusto.com slash twist. And Nexos.ai. Stop shadow AI in its tracks with the unified platform for secure AI adoption and productivity. Try it with a free 14-day trial at Nexos.ai slash twist. That's N-E-X-O-S dot AI slash twist. Hey, everybody. Welcome back to This Week in Startups. This is Alex. And today on the show,

SPEAKER_06: I have three amazing Twist 500 companies coming your way. First up, we are going to talk to Medium CEO, Tony Stubblebine. Now, if you're a long-term listener on the show, you know we've had Tony on before. So why now? Well, just last week, Medium, along with other companies like Yahoo and Reddit, all joined together under a new initiative called RSL, or Really Simple Licensing. Think of it as the RSS for the monetization of media in the generative AI era, if you will. I'm curious about why he picked this project back and why now? Then after that, we're going to talk to Human Native. Now, this is a Twist 500 startup that I added originally to the list because they were building an awesome AI content marketplace. Essentially, publishers and AI companies would meet in the middle under the startup's auspices and find a way to make a deal. Well, it turns out they're pivoting a little bit. I learned a ton in this chat about the pitfalls, you might say, and the potential in monetizing IP in the generative AI moment. Then, to wrap up our show today, we're going to talk to the CEO of Tello Trucks. Now, if you recall Slate, the Jeff Bezos-backed small EV truck company, this is like that, but cute, and also pretty far down the road. So if you care about EVs, you care about transport, well, Tello Trucks is an absolute treat, and I want one. But no more from me right now. Let's dive into it, and here's Medium's CEO. All right, now we're all big fans of

SPEAKER_09: using AI search tools. I'm a big fan of ChatGPT's GPT-5, but you might be a perplexity guy. You might be a Claude fanatic. You might love what Google has cooked up. One thing that people don't really understand is how many queries these search engines send out. They might read 10, 20, 30 pages, and what that has led to is a change in how the internet functions. Now, a lot of websites are seeing an enormous number of pings and crawls from AI search engines, AI search agents, and often the compensation structure for those is zero. So companies are working to find a way forward to ensure that there's good attribution and good economics on the side of people who write content. Cloudflare's paper crawl model was one such effort, and we have a couple startups out there, things like Tolbit and Human Native and Created by Humans, that are building their own marketplaces. Now, this week, there was a new product that came out called Really Simple Licensing, or RSL. If you know RSS, well, it's a term that you're probably pretty familiar with, and it's a fun little play on that old school framework. One of the companies that's signed on to RSL is Medium, along with Reddit, and Yahoo, and others. I'm fascinated by this new approach to ensure attribution and compensation for the internet and the world of people who scribble online. So please join me in welcoming back to the show, Medium CEO, Tony Stubblebine. Tony, how are you? Hey, I'm good. Thanks for having me back. Dude, my absolute pleasure. So catching people up a little bit, last year, Medium hit 1 million paid subs in April. You announced cash flow positivity in May of last year. Big steps for the company. And then since then, Tony, AI has only become more popular, and we've seen the rise of generative AI search, and as I mentioned, a lot more queries. So just to set some context here,

SPEAKER_10: how much more now is Medium getting hit up by AI search engines than it was last year when you reached

SPEAKER_00: those financial milestones? Yeah, I mean, the rise of AI sort of impact is everywhere. Like, we see it in the slop that gets posted to Medium, which we do our best to either delete or to hide. You know, I think we're more focused often on hiding than anything. We see it in the comments, and we're like constantly fighting inauthentic comment bots, and we see it just in the traffic to the site through all of the crawlers, because it's not just the AI companies you've heard of. It's all the up-and-coming wannabe AI companies that are trying to get new training data. And I think what we found is, you know, one of our jobs is we have to represent the writers on Medium. Like, without them, we're nothing. And in particular, we're fighting this issue with the AI companies, which is not technological, it's human, right? I think they, the way that they have operated to date breaks the social contract, that they, you know, essentially society is based on an exchange of value. And they've taken value from the writers and creators of the internet, and really offered nothing back. And that's not good, it's not fair. I think a lot of our writers would call it theft, and like literally do. But more, it's like, I mean, we could just think of this as an urgent internet need. If that exchange of value goes away, the public internet goes away. This is what people are calling the dead internet theory. It's just going to be only AI slop, and all of us creators and writers will kind of return to private spaces and to paywalls, which is like, I mean, there's some something to that, but it's not the end of the world. But I think, you know, if we want to protect a public information superhighway, we have to fix this thing. And the core thing we're trying to fix is the behavior of the AI execs and to get them to come to the table and find some workable model

SPEAKER_09: that works for all parties. Yeah, and that's why the really simple licensing idea resonates with me, because it's not one company going out there and saying, this is what we're going to do. It's a collective of very highly trafficked websites and companies. So I guess take us back to the beginning. When did this idea come to Medium's shores? And how did you decide that it was the right approach, amongst others, for Medium to take?

SPEAKER_00: Yeah, we decided it was the right approach before we knew of this particular approach. Right away when ChatGPT launched and we started to see what was happening with kind of the lack of exchange of value, like the theft, essentially, of people's content, that we thought, like, there's no way to resolve this except by force. Like, if they're going to be antisocial, we have to be antisocial. And the way to do that is to get a coalition going. And I tried to get a coalition going. But we're not really quite, you know, we're not Facebook or Meta, as they call themselves now, right? We don't quite have that heft. And I think what the bigger companies tried to do is just cut individual deals instead. And so if there were individual deals to cut, those have all been cut. And now we're at the point where what's left is the coalition. So it feels a little bit late. But overall, I'm happy to have people going in the direction of an internet standard. Like, we can't fight what's going on with vendor lock-in. We can't fight it one-on-one. But if we get together, we have enough clout to say, look, you're not going to be able to train on new data again, unless you come to the table. Or worse, you're going to be the individual company that's not able

SPEAKER_01: to train on new data, while these other companies, while your competitors do, because they pay for it.

SPEAKER_00: And so that standard, when we saw it, we saw it pretty early. I think we were one of the first companies to sign on. And I bet you we were the fastest. Like, between seeing it and saying yes, was like, probably 30 minutes. Which is instantaneous in business time. Yes. And it's because we'd already thought it all the way through. Like, we knew what we were looking for. And we had been vocal. And I think that's also why they came to us early. I said, look, you know, one of the creators, Eckert Walter, is also one of the creators of RSS. So like, I understood he was a credible creator of internet protocols. I worry for your listeners that here we are talking about internet protocols. But there's something exciting in it for me. And so we signed on. And really, the only thing we changed is that we knew that our intention was, if we negotiated with AI companies, that we would pass all of the money back to the writers. I think we're the only- All of it. All of it. Okay. My CFO is like, Tony, what about the legal fees? Stop saying all of it. Okay. All right. Caveat. Maybe if the legal fees are really high, we're going to pay, like, get it, like, at costs, let's say. We're going to pass it all through to the creators. We already, like, Medium's just

SPEAKER_01: not in the cellular data business. We never have. This was, again, an easy decision. Doesn't even matter how

SPEAKER_00: much money it is. It's a pass through to the writers. And as far as I know, we're the only platform that is thinking that way. And I mean, I was like, along with shaming the AI companies, I'd like to shame the other platforms that they also should be doing that.

SPEAKER_25: There are some amazing AI tools out there that will absolutely make your workers faster and more productive. But having a large team using all kinds of different AI tools, sanctioned or unsanctioned, can actually pose some pretty serious privacy and security concerns for your data or your customers and partners' data. This is shadow AI, and it could be costing you millions of dollars, and it could be exposing your company to risk. But there's a solution. It's nexus.ai. Their workspace gives your team a secure browser-based environment where you can work with the latest and greatest tools and models while giving your admins and security team full visibility and oversight. What if somebody on your team says, analyze all this compensation data? And then it winds up training a model on those people's compensation in your company? You want to be compliant with all your policies. You want to protect your data. But don't believe me, try it for yourself. Go to nexus.ai slash twist for a 14-day free trial or check out the link in the episode description below. I love having you on. Also,

SPEAKER_09: I think people are interested in internet protocols because they're the framework on which the internet sits. And most people that watch this show are building internet-based companies. Some are doing hardware, absolutely, but mostly it's online. So I think this is pretty down the pike. Now, you are talking about AI training, and I had framed this mentally more as an AI inference point because RAG queries go out there and ping, as I said, dozens of sites. It's fun to watch Google's AI go, ping 70 sites. I'm like, sweet. But that puts a big load on things. So I had it more framed from the

SPEAKER_10: inference side. You're talking more about training. Is that a better way to think of things in this case?

SPEAKER_00: That we're talking about defending training data. This is, I mean, this is where you've like fallen into the complexities of it. I mean, certainly training is one because there is no exchange of value, not even sending traffic back. The RAG side of it that you're talking about, where they make a summary, essentially the AI-generated summary world, often has citations and often sends traffic back. So at least they're creeping back into the world of an exchange of value. Problem is the exchange of value is very weak. Like, you know, I would say it's like, to the degree that Google is trading their prior search traffic with like a citation and a generated result, it's probably we're giving up

SPEAKER_01: like 100 clicks to one, you know? And so it's just, it's minuscule. And it's not nothing. I mean, I will say that the traffic that comes from ChatGPT now converts to a paying member on Medium four times higher than normal traffic. It's higher intent, as you would expect. It's like a person says, hey, this summary is not enough. I want to read deeper. Oh, that's our dream Medium reader. Like, we like people that think deeply and read deeply and care about being smarter and understanding

SPEAKER_09: all the complexities of something. But higher willingness to pay doesn't imply more total pay if the amount of people coming back to Medium is lower. So does it net out to be

SPEAKER_19: even? No. Or is it still dramatically? Okay. Not even close. Got it. And so, and that's the fear,

SPEAKER_00: right? Is that what we have is really like a temporary moment where our business is shaky,

SPEAKER_01: like, you know, the sort of the social media platform business is shaky. And also the businesses they're building is shaky. Like what's going to go happen to the world of rag results when there is no more, you know, rag to retrieve, right? Right, right. No, that's what I've been thinking

SPEAKER_09: about. Like the best thing possible here is that if there's a way to bring monetary value from the AI companies to the creators in a way that works for everybody, then we'll have healthier creators, short term and long term and also healthier AI companies long term. Like, like, it feels like we're shouting for all the systems to work at once, which feels a little bit surreal. Right. And so

SPEAKER_19: what's weird, right, is these are smart people on the AI leadership side. Why didn't they see this

SPEAKER_01: coming? Why do why didn't they start with a collaborative approach? Right? I mean, like, as I said,

SPEAKER_00: they started with an antisocial approach. It's like, grab it and then come fight with us. And this is where I think Matt Prince at Cloudflare actually like had it right. It's like they have not voluntarily come to the table. So the first step is to force them. And that is a mass blocking of

SPEAKER_09: crawlers everywhere. Has that had an impact, do you think? Because me, the Cloudflare has been a little bit quiet about paper crawl since it's kind of thunderclap announcement. And I don't have a good feel for how impactful that's been. And also, Tony, I'm curious, did you guys consider that before

SPEAKER_19: going the RSL route? Yeah, we have two problems with it. But I say this with a lot of respect for

SPEAKER_00: Cloudflare. Like, we needed people to move early, and they did, and they moved in an articulate way, and they moved in the way that they were most capable of doing. But at the end of the day, you can't protect the internet through vendor lock-in. You know, we can't say, oh, let's all sign up for Cloudflare in order to solve the problems of the internet. And so at the end of the day, we're going to need Cloudflare to support RSL. And then this is this particular thing for us, which is we made an addition to the RSL standard that would allow us to do the block on a per page

SPEAKER_01: basis. And Cloudflare does it on a per site basis. And that's because most companies are viewing this as we want our company to get paid. And we're viewing this as we want individual people to get

SPEAKER_00: paid. And so we need to then give those people their own access controls, because a lot of them

SPEAKER_01: are not going to opt in under any conditions, because they're so morally distraught with the

SPEAKER_42: AI companies. Yeah, there's a lot of folks out there who are, and there's a lot of folks out there

SPEAKER_09: who are not. And I think there's a lot of folks who are in the middle who are probably your target here, because some people are like, screw it, crawl me, I don't care. And some people are, you know, very much on the other side of things. Okay, so we both agree that having a unified front here, a united front is super important. And I think with the other sites that you mentioned, or that I mentioned, there is enough heft there to make this stick. So what's been the response to your knowledge from the AI side of things? Because we can talk about the publisher side until the cows come home. But if the AI companies don't engage, I wonder if this will work. So any encouraging

SPEAKER_00: signs on that side of the fence? Yeah, it's I think we're going to find out. It just launched. I happen to have already something scheduled the next day. And but they were not anywhere near prepared to speak to it. But I think this idea that first of all, we have to show credible force. I just, I don't like this is not how I like to do business. But this is kind of, you know, like this is the prisoner's dilemma. Like, if they like, how they show up is how we have to show up. And so credible force is to have enough content kind of under this umbrella. And I think legal threats have not really worked. Can I tell you a lesson that Medium learned early on? I would love to hear it. So it really early on, we realized that even though like that, we're a big enough data set, that we're able to poison any language model. And the reason we learned this is because Medium, like kind of by design, is just filled with em dashes. So you know, this theory right now that you know, it's AI generated because it's filled with em dashes. Well, you know where that came from? Medium, because our founder loved em dashes, popularized a feature, like an automatic feature in the editor that would convert various dashes and whatnot to em dashes. And so then it's became culturally a way to write on Medium. And it did spread beyond Medium itself. I think most modern text editors on the internet right now do this, like if you do a double dash, you'll automatically connect. But you know, essentially, we created a trend, which now shows up in the language models. And so like having seen that, whenever someone comes to us and says, you know, like, look, you know, you don't have legal standing, how are you going to block them? They're just going to get around, like the crawlers will get around you somehow. It's pointless and whatnot. I say, well, you know, like at the end of the day, we could go back to just poisoning our results. Like, you know, I don't know, like the silly way is like, oh, you know, it really increased Medium's Riz if we got involved in like slang maxing or, you know, the results. Like we can rewrite like whatever gets returned to the crawler with whatever crazy modern slang like we want. Or we could get really, you know, like we could get a lot harsher than that. Like, you know, we can essentially put slander into the results. Like anytime you hear the word open AI or see the word open AI in a Medium text, when a crawler is reading it, we can just say comma filled with hallucinations, comma, and just move on. Right. And so now like that's every, like every language model is filled with the slander. And it's like, hey, that's your fault. Follow our terms of service. Right. Yeah. So I think, you know, we'd like to avoid that level of warfare and instead just like, like deal with reality. Like, like these companies don't get

SPEAKER_19: this for free. And if they don't do something, they're like the whole, the whole foundation

SPEAKER_25: they're built on will disappear over time. We talk all the time on this podcast about the importance of moving at startup speed. Being a founder is about prioritizing and managing your time and you got to be ruthless about it. Well, thank goodness for Gusto. We love and use Gusto. They're the online payroll and benefit experts. So I don't need to be. It's all in one remote friendly and incredibly easy to use. So you can hire on board pay and support your team from anywhere. They offer so many helpful automated tools and features. Everything you need is templated and built right in from offer letters, onboarding materials, setting up direct deposit and more. It's easier than ever to switch your team over to Gusto, G-U-S-T-O. Just transfer your existing data to get up and running fast and you don't pay them a cent until you send your first payroll. Try Gusto today at gusto.com slash twist and get three months free when you run your first payroll. That's very generous. One more time, gusto.com slash twist. And again, write a micro review if you love Gusto and you found out about it here on This Week in Startups.

SPEAKER_09: So you said in your post on Medium's implementation of RSL that you're doing, quote, the simplest version of this new RSL standard, which prohibits AI companies from using your stories to train their AI models, but allows them to summarize and link back to your writing in AI generated search results, effectively closing the training door and demanding fair attribution on the what I would call the inference or rag side of things. The rag side, yeah. Seems like a good, a good starting point. When do you think you're going to have the either demand from the AI side or the confidence to roll out more of RSL and essentially start executing on paper

SPEAKER_11: inference? Yeah. We are not going to make any change unless an AI company comes to the table with

SPEAKER_00: money. So in a way, the announcement from us is a start of a discussion with our own community about how we're planning to negotiate on their behalf. I didn't want to do this in a way that wasn't transparent to our community. But now that it's out, we can literally start having these conversations.

SPEAKER_01: And the conversations kind of don't make sense until you have a number, right? Are we saying we're going to pay every story one penny? No one's going to care, right? Is it $5? Well, some people are going to start to care. Is it the best stories that show up in AI results over and over again? Are they going to make a hundred, a thousand dollars? No, that's going to catch people's attention. And so, you know, we have to start the negotiation to see where it goes. You kind of trod on my next question there a

SPEAKER_09: little bit, which is what do you think the economics of this could become in time? And I don't know what I would ask this, Tony, because the answer is going to be, it depends. But let's say that I'm someone who has written a thousand pieces of content on Medium over the years. So I'm probably one of your power writers, if you will. And my material gets reasonable out of search traffic. I have some dedicated readers in the app, I have some paid readers, and there's also this AI component. So do you think this replaces a material percentage of subscription fees in terms of like how much money a writer might get? Or is this more of a dust on top? I'm trying to figure out the scale of possible

SPEAKER_00: return for your more active writers. Yeah. I think that the standard internet pay rate is probably where we should end up. And so, you know, you could look at this through the ad model, right? If we're, if Google sends you a thousand clicks and you have like run of the mill ads on it, you can make around $5, right? So the $5 CPM, $5 RPM, like that is roughly the standard pay rate. And some people do quite a bit better than that. And that is kind of the business of media these days is to surpass that. But your average person is, is going to be in that, in that world. So you've got an apples to orange thing where they're not going to send a thousand clicks because literally the summary is stealing your clicks, but they are going to have to give something in that, in that realm to even be on par. And if they go below, you have to factor in, they're also not sending you the clicks. This, like this thing that got you to write online in the first place, this validation of having people read you and this validation of having your ideas spread, that's also going away. So I would say like, in order for this to actually work, to actually create a healthy ecosystem, they're actually probably going to have to come up from that, that level.

SPEAKER_09: I want to close with just a question about AI and writing in general. You joked about the M-dashes and how you changed the way people think about AI writing. By the way, screw you, because I'm a big M-dash guy. And now people are like, oh, AI wrote this. I'm like, no,

SPEAKER_35: these are artisanal hand-grown words, my friends. AI trained on the best writers on the internet, and now it uses a lot of M-dashes. That is how it happened. You're welcome, everybody. You're

SPEAKER_09: freaking welcome. Now, my concern is that people now use AIs to generate summaries of writing, and they often use AI to create words. And there does seem to be people slowly backing away from the process of writing things down and reading them. And of course, technology changes, people's habits. I'm not here to be a Luddite, but I'm curious if that represents a material challenge to the ethos of medium, which is writer first, writers writing, and people reading those writers. Is the culture shifting away from what medium does? I'm always shocked. We get so many questions

SPEAKER_00: along the lines of what you're saying of culture, trend, whatever. And I just come back to,

SPEAKER_01: writing is thinking, and reading is learning. For thousands of years, humans learn through story for a reason. The way our brains work require all that context. And the idea that summary is going

SPEAKER_00: to replace story is kind of ridiculous and not even a new concept. The cliff notes didn't replace the books. And so I always come back just to the first principle. Does the world still reward people who want to be smarter? Yes. Does your life get better if you're smarter? Yes. So in the world of

SPEAKER_01: smart people, writing makes you smarter, and reading stories from other people makes you smarter. And so the medium business is built on that premise. Reading and writing is for smart people, and there's always going to be people that value being smarter. Well, if people want to learn more

SPEAKER_09: about RSL and what it is, they can go to rslstandard.org, and of course, medium.com for all things medium. Tony, as always, an absolute treat. Once you corral a couple of AI companies by the neck and drag them to the negotiating table and get them to slap their checkbooks down the table, come back on and tell me how that goes. And I'm going to be watching very closely to see how the economics work out for the medium writers out there, because viva la writing. It's too important to let die. We're both fans of writing. Thank you. All right, Tony. Bye.

SPEAKER_76: So continuing our conversation about intellectual property in the era of generative AI,

SPEAKER_78: we're next speaking with a Twist 500 startup based out of the UK that was building an AI content marketplace, a system in which IP holders could license their content relatively easily and quickly to AI companies that wanted to pay for it. However, the company is pivoting a bit. They have new plans for the future. So we're going to talk about why they ran into a dead end and what they're going to do next. Please welcome to the show. It's human native co-founder and CEO, Dr. James Smith. James, how are you? Hey, Alex, I'm good. Thanks. Thanks for having me. Yeah, my pleasure. So I had put your company amongst several others, your toll bits, you're created by humans as a company working to connect IP holders and AI companies. And my view of this was always that there was going to be ample supply that publishers, authors, people who own datasets would love to license their data and get paid for it. You just told me the company is pivoting a bit. So I presume that on the demand side, there was a bit of a mismatch. Is that fair?

SPEAKER_82: I think there was a mismatch, but I'm not necessarily sure it was always just about the demand itself as to,

SPEAKER_85: like, it felt like a communication breakdown a lot of times, right, frankly. So you're right. I mean, rights holders want to get paid. They're facing immense pressure. Their traditional business models on the internet are collapsing. We need a new economic model for the internet around AI. But rights holders think about content as content, as their life's work. It's very emotional. They put a huge amount of effort into it. And AI companies think about it like data. It's numbers that help them improve their models. And that is a real fundamental mismatch. And so what we were finding is that rights orders would come and talk about, you know, the respect for their work and having things like editorial control over headlines that were reproduced. And AI companies would come in and say, like, okay, that's great, but we want this really specific requirement. We want exactly this and nothing else. And we're going to pay for just this. And what's the ROI? And can we prove this is going to improve our model performance? And so it became this quite difficult conversation

SPEAKER_87: between two sets of people who didn't speak the same language. We attempted to play translator,

SPEAKER_78: and that was really hard. So James, I'm really curious about the difference between data and content. To me, it's tomato, tomato, but you're making it sound like a really big difference. So is this a structure difference? How the information is presented? Is this a type of data question? Or is this more of a philosophical difference in just how people think about the

SPEAKER_90: information in question? I think a lot of it to do is to do with structure. A lot of it's to do with

SPEAKER_85: rights holders have been creating content over decades, perhaps. And for the last maybe 10 or 15 years, they've been throwing it into a cloud storage system. And they might not have a good understanding of what they have. And, you know, we worked with a rights holder who'd been approached by an AI company asking for Arabic language content with transcriptionists. And they were like, yeah, we're pretty sure we created some stuff like that, but we have no idea where it is

SPEAKER_87: or what's in there. And they were like, if you can find it in our archive and here's access, then good luck. You can, you know, you can help license it. That's a real challenge for them.

SPEAKER_94: It's not always easy to get your management and engineering teams to see eye to eye. But now to

SPEAKER_25: bridge this gap, there's DevStats. You got to check out what the team over there has done. They've built a tool that translates complex engineering metrics into a shared language everyone at your company can understand and helping you spot bottlenecks and most importantly ship 30% faster. Plus I love this. You get individual contributor insights so you can instantly see and understand each team member's performance. We talked to one team. They discovered that nearly half of their developers time was going to low impact bug fixes armed with this info. They were able to revise their roadmap to focus on high impact work. And now they're shipping new features twice as fast. Sometimes within just a couple of weeks. So stop shipping late. Check out DevStats today and get 20% off plus access to their dedicated Slack channel. That's a really good idea for a startup. Have that Slack channel so you can talk directly to your customers. Go to devstats.com slash twist for your 20% discount.

SPEAKER_97: But the philosophical argument is also huge. You've probably seen there's a lot of debate happening

SPEAKER_85: both in the United States and especially in the UK around this idea of fair use, or we don't have that in the UK, but the ability to use this content for copyright. I've been lucky enough to be asked to give evidence at the House of Lords and the House of Commons select committees on these topics and met rights holders. And for them, it is a really emotional issue. That philosophy part of it of this is my life's work. How dare you copy it and change it and then use it for your own purposes to make money. That is a really fundamental challenge for them. And so the way that manifested themselves is sometimes in the business aspects, they would come in with very different ideas about price and tone and level of control. And as I said, we attempted to translate and be the buffer between

SPEAKER_78: those two. So publishers wanted high price, low shared control. And I presume on the AI side,

SPEAKER_10: they wanted low price, high levels of control. And that would probably sound like a different language at the negotiation table. Exactly. So how long did you guys try to bridge that gap? Like how long did

SPEAKER_104: you sit there and be like, okay, we can make this work, we can get Bob and Jane to talk it out and reach a common understanding before you were like, okay, this is not going to be the path forward.

SPEAKER_97: So we did this for about 15 months, we started our company in April of last year, and we went on until

SPEAKER_85: about the end of q2 this year with a marketplace model, we attempted to prove that we could connect these two parties. We then tried to do a few different things. And one of the things that we tried to do is, I have a very clear idea of what human native the marketplace would look like if the company was a billion dollar company, it would be we'd have to solve the ability for trillions of transactions happening effectively in real time for fractions of a cent each time. And our company would look like infrastructure, and we would be that low friction barrier to enable these two parties to transact in close to real time. But in order to do that, you need a lot of precedence, you need a lot of standardizations, and you need the ability for, you know, supply and demand to meet in that way. In the end, it felt like we'd started out being this buffer between the two, we tried to let them speak to each other, and then provide the infrastructure and the tools to do that. And eventually, it just felt like being involved in licensing was not the right path forward for us as a company, and frankly, wasn't a good use of our talents. And a couple of things, I've been a product manager at Google and DeepMind. We built lots of technology as a team.

SPEAKER_87: I'm not entirely sure I'm the world's best salesperson, or at least the best middleman between two organizations that are trying to get the best contract.

SPEAKER_106: Those are different skills, who would have thought?

SPEAKER_85: Yeah. And so I think it's time for us to focus on what we're good at, which is building product and building technology.

SPEAKER_78: So we'll get to that in just a second. But I want to go back to your point about trillions of transactions for a fraction of a penny, because there's two main ways that data is used by AI companies. And a year ago, we were talking mostly about training data, AI companies going out there, hoovering up a large chunk of the internet, and then using that to form the large and their large language models. Lately, though, talking to both publishers and entrepreneurs, it seems that the RAG use case, using data real-time to help serve a query, has become much more important. So as things go from kind of the training use case to the RAG use case, do you think that's going to help to resolve any of the issues that you've mentioned? Or is it just the same problem as before, slightly different application, but no change to the fundamental disconnect that we just discussed?

SPEAKER_109: I think it's still the same fundamental disconnect, Alex. I think the challenge is going to be,

SPEAKER_111: do publishers want to participate in a RAG-type system where they don't get any control over the user experience, and they lose their direct relationship with the customer? The thing that's changing about these generative AI platforms, people are not leaving them.

SPEAKER_85: They're walled garden systems, and you've seen traffic referrals drop off a cliff, and so if you're a media publisher, you don't have any incentive to work in one of these systems, other than perhaps get a little bit of revenue, because you're losing your direct relationship with the customer, and your ability to control how your message lands. For the AI companies, their situation is going to be, well, why should we pay for this information if it's available on the free internet, on the open internet? And I think that's a really difficult one. The other fundamental thing, and the thing that's dissuaded us from building our own RAG solution, because we had a working version of this, was simply that AI companies have that, a lot of the time, it's not invented here syndrome. They want to control critical parts of their infrastructure, and that's really understandable. And so if RAG system and working with publishers is going to be a key part of how they deliver their user experience, they're going to want to own the technical infrastructure that makes that happen, and they're going to want to go out and talk to publishers and bring them into that

SPEAKER_78: platform independently. Yeah. So if the data structure and discoverability problem hadn't existed, do you think there would have been a way to translate between content holders and AI companies? Or would the other issues that you've mentioned be enough to still scuttle that as a

SPEAKER_87: possibility? Well, I think the media organizations and large tech companies have had a storied history

SPEAKER_85: to this point. You know, there's 20 years of history in some of these conversations. And so getting into the middle of that as a startup was also a challenge, like quite frankly, right? And so I think there's

SPEAKER_111: a lot of distrust that even if we can solve all the technical issues, there's still some fundamental

SPEAKER_78: barriers to overcome. This does bring up one of my favorite quotes. In 2024 coverage of your seed round and tech runs, you said, I'm the CEO of a two-month-old company and have been able to get

SPEAKER_10: meetings with CEOs of 160-year-old publishing companies. And I was like, well, yep, that's the power of tech opens a lot of doors. Okay. So let's talk about solutions then. Because it sounds like right

SPEAKER_78: now the idea of building a shared place for demand and supply to come and meet and reach an agreement

SPEAKER_10: and then execute it is a bit rough. So what do companies need that you can bring to them soon

SPEAKER_78: to help them take their information and either use it internally for their own AI usage or make it available for sharing? Because it sounds like you're leading me towards a, there's a technical solution to part of these problems and we're going to go build it. Well, that's what we're hoping. Yeah. I mean,

SPEAKER_121: obviously I'm an entrepreneur, so we're trying to sell a vision about what we believe in,

SPEAKER_85: but definitely this is what we're seeing. When we, we, some of the partners we work with and we have worked with some very large publishing partners are doing their own deals and have been successful. I've made tens of millions of dollars in AI licensing deals. You can say, you can say pounds. It's fine. I'm just being respectful of your, you know, majority of your audience. Also having worked for Google, sometimes I'm kind of this mid-Atlantic half English and half American in the way that I speak. Just don't say kilometer

SPEAKER_111: or France and we'll be fine. I'll just do it all with the Scottish accent and then nobody will understand what I'm saying anyway. Um, so actually as a fan of still game, the Scottish

SPEAKER_124: television show, I think I'll be okay. You've seen still game. That's great. I always recommend

SPEAKER_78: that. That's a great way. Every single, every single episode several times, including the more recent seasons, which was good, but yeah. Oh, and I've seen the stage shows and the original

SPEAKER_125: stage production. So yeah, you're here. You're a mega fun. Wow. Wow. Okay. We should talk about

SPEAKER_111: that later. Um, back to you. Yeah, back to, uh, dollars. And so the, the challenge I think is going

SPEAKER_85: to be helping these companies answer the more sophisticated AI buyer requests. If an AI content company, AI licensing company comes along and says, Hey, we want to license only videos of sunscreen bottles with busy backgrounds, not clear backgrounds. So Alex, not James in this scenario, how do you, how does a content company that doesn't have any particular expertise in building AI models

SPEAKER_127: answer those questions? Well, it sounds like what they would need to buy is a third party AI service

SPEAKER_78: that would go through all their information and crawl to find it. So that sounds to me almost like an internal search engine and taxonomy generator, something along those lines. And that's the kind of

SPEAKER_121: types of pilot programs that we're doing at the moment. We're helping companies make their data

SPEAKER_85: assets useful. Because if you think about the last 10 or 15 years, there's been a huge amount of progress in big data. There's lots of category defining companies, which have helped organizations make use of data. Snowflakes, Databricks. Yeah. Palenters, which help people basically make use of data and do things with it. What happens if your data in that question is images, video, audio? I don't think there are category defining companies yet in that space, which help you extract in meaning and useful value out of that content. That might be that value might be external. It might be for licensing opportunities, but it might be internal. One of the companies we're speaking to has a lot of call centers and they have a lot of audio recordings from their call centers. They would like to be able to analyze those and extract meaning. What are their customers saying to them

SPEAKER_87: through those call recordings? I think that is a really interesting space. And there's a lot of

SPEAKER_10: possibilities there. Does the system that you're envisioning to help people find videos of sunscreen with busy backgrounds versus non-busy backgrounds and to dig through call center data, which I presume is transcripts and call links. So it's both words and data points. And audio. Oh yeah. And audio. Do you

SPEAKER_78: use traditional non-AI technologies to parse that information in this vision? Or do you use

SPEAKER_10: anything predicated on generative AI? I want to ask about vector search, if it's the latter, and if it's the former, I don't. So just guide me with how you think the tool is built.

SPEAKER_85: I think it's a bit of both, quite frankly. I think there's a lot of great software engineering and data engineering of, you know, traditional means, but now that is more possible at large scale. And then I think, yes, I do think generative AI has unlocked a lot of ability to understand content in particular. And so there's some really interesting techniques that we're exploring there.

SPEAKER_78: So is this a system by which companies can prepare, or is it a system by which they can explore? Because to me, we talked about, you know, the lack of data being searchable and findable. So you could help people find it or help them kind of structure it in a way that other people can look into it. So I guess, what's the product that you're pushing towards here? And what does it look like when

SPEAKER_85: it reaches the market? Oh, you're truly pushing me at this point, Alex. This is what we're trying to experiment with right now. I think the two are fundamentally linked. I think preparation of content enables exploration. And the goal might be exploration or use, but you can't do it without the preparation. And so we are, for example, working with a company today where we're going to be showing them the latest version of what we built, which is an interface for understanding and searching their content, but it's all built on the work we did to prepare their content.

SPEAKER_83: I see. Okay. Now, one of my favorite companies, Box, because I've talked to the CEO,

SPEAKER_10: Aaron a bunch of times and he's just charming, has been talking ad nauseum for the last couple of quarters, maybe years at this point about helping people who have their data stored inside of Box, which is a traditional enterprise. Oh man, it's been a while enterprise sync and share files, EFSS. I forget the acronym, whatever it is. They have a bunch of data in the cloud for their customers. And so they're building tools that I think are kind of aligned to what you're describing because they already have the data. So why not help people kind of figure it out what it is and use it. Would your system work best for companies that have a lot of on-prem storage, or do you think this is

SPEAKER_85: a multi-cloud affair? We are trying to build this solution where it doesn't matter where the data is. So the data could be on multi-cloud, the data could be on-prem. It's probably for people who, sorry, Aaron, the data is not already in Box because there's a lot of people who have been using lots of different systems to store their data. It could be, they've been sticking it raw into S3. It could be Dropbox, it could be Box, it could be Microsoft SharePoint. It doesn't really matter where

SPEAKER_113: the data is, but people need a unified view and then they need to actually do something useful with

SPEAKER_78: that data. Okay. And then on the do something useful with that data, is that a thing that you think

SPEAKER_83: your company is going to focus on or more like help people get to that point? And then from there,

SPEAKER_87: it's choose your own adventure. I think it's honestly, we're still early enough that we get to figure this out. And that's the fun bit. I was talking to a very senior person at a multinational global company. Sorry, I don't like using names because we're under a lot of NDAs. No, no, you're fine. That could be one of like 100,000 people.

SPEAKER_85: Yeah, exactly. This was like a C-level person at this company. They have huge archive and they're like, well, yeah, we could license our content for 10 million a year. And then somebody will build a

SPEAKER_87: billion dollar product on that content, but we'd like to have a shot at building that billion dollar.

SPEAKER_78: Oh, interesting. So you're facilitating these companies that would have come to the marketplace

SPEAKER_83: for a cut of those, you know, trillion rag calls at a fraction of a penny each and instead allowing them to build something on their own. Exactly. Because what I said to him was,

SPEAKER_85: well, have you figured out how to unlock your archive yet? And they're like, oh, no, no, we haven't. And I was like, cool. That's where we can help.

SPEAKER_10: So a question about that, because the way that I've always seen this is

SPEAKER_78: data in aggregate is powerful. Data in smaller chunks is less powerful. But I'm also aware of

SPEAKER_10: what you said earlier about the sunscreen example, which sometimes an incredibly narrow slice of data is very, very powerful. But that's all from the perspective of these large AI companies, your Anthropics, your OpenAI, your XIS, et cetera, your Mistrales. For companies that would want to build something on top of their own data, they only have their own box at that point, right? So in this case, the publisher has their list of content through time. Is that a broad enough data set? Or is it perhaps a narrow enough data set to build something useful on top of it? Or will most of that value, do you think, come to fruition when other companies can combine multiple data sets from similar companies to build something? Because to me, it's cool that they want to build a billion dollar company on top of it. But do they have enough grist for that mill?

SPEAKER_85: It's a really great question. And I think there's so many possibilities now with these advances in the models that we're seeing. If you were to take a great third party model or open source model and then apply your archive to it and create an AI product, which deep dives into your article and gives you an insight into that. If your articles, if your archive is rich enough, that's a really cool product. But equally, you can also take your open source model, your archive, and then as we just talked about, use a rag system to bring in the content you don't have. And then there's also an equally compelling product. This is a choose your own adventure story. And I'm really excited to see where it goes.

SPEAKER_10: So the Guardian, to pick a paper that you and I both read, could take an open source LLM, bring its own data after working with your company to get it all suited up and booted up to go. And then they could also rag out to, I don't know, the times, pick a times from either of the side of our ocean, and then have an expanded Guardian AI model that can help explain news even. Okay, I can see that. That's going to

SPEAKER_78: require a lot of technical leadership at companies that have famously not been so technically leading. And I'm making fun of my own industry here, journalism, just to pick one. Do they have enough

SPEAKER_76: chops to do that, James? Or are they? They're bad at websites. Okay, like, how can they build AI systems?

SPEAKER_138: Well, I think we're going to find out because I think the barrier to entry to these systems is

SPEAKER_85: getting lower and lower, right? Today, it's still quite hard for a non-technical person to use some of these AI systems in a way that's not just using ChatGPT. I don't know if you saw, there was a great podcast. Sorry to talk about other podcasts on this podcast. There's a great podcast that the leader of chat PRD did with Tom Tungas from... Oh, Tomas! Yeah, Tomas, sorry, from Theory Ventures. And he talks about how he wrote a script to take his 36 podcasts that he subscribes to because he doesn't have 36 hours in the week to listen to them all, to then extract insights from them so that he can get a quick digest of like what's happening in tech this week. And I think it's a really cool example. But because Tom is incredibly technical, he was able to write that script and do it. What if there was a system that enabled many more people to do those types of things? And so we talk about

SPEAKER_87: technical leadership at these companies. What if it's not technical leadership? What if it's just

SPEAKER_78: business and operational leadership? Ah, and so the technical leadership would then pool at a company, perhaps one based out of the UK, perhaps one called, I don't know, Human Native? Human Native AI. That

SPEAKER_06: would be nice. So James, just before you go, drop your URL and a job you're currently hiring for.

SPEAKER_85: We are humannative.ai. That's one word, humannative.ai. And what we're hiring for will actually surprise us. If you think you can provide value to what we're doing, I think we're really interested to speak to you. We're looking for high agency people who can help us figure out what's next. This is a really interesting market. There's a lot of good to be done. I'm excited to see where it goes.

SPEAKER_173: All right, James. Thank you very much. Thank you, Alex. Cheers.

SPEAKER_09: Hey, welcome back to Twist. Now, I learned to drive in an F-250 stretch bed with an enormous gear shift coming out of the floor and more torque than a tank. It was a great car to learn to drive in

SPEAKER_06: because you basically couldn't stall it thanks to how grunty it was. But since my youth, trucks have gotten bigger and bigger, often without any increase or even a decrease in their cargo area. In short, I think that most American pickups today look and operate more like a minivan with a wheelbarrow attached. But there are a couple of companies out there who are thinking differently. So what if we made smaller trucks with actual functional beds and maybe batteries instead of the two huge fuel tanks my dad's truck still has? Tello is doing just that. And to tell us more about that, please welcome to the show, Jason Marks, co-founder and CEO of Tello Trucks. Jason, hey, how you doing? Hey, thanks for having me. My absolute pleasure. So your company is building something called the MT1, which I have to say two things about. One, it's very small as far as trucks go. And two, it's absolutely adorable. But why don't you tell us about the truck and why you picked this particular form factor to start with? Yep. So we build mini trucks. That's what the MT and MT1

SPEAKER_179: stands for, mini trucks. We build crew cab pickup trucks with the same capabilities as a mid-sized work truck, like a Tacoma, maybe a smaller F-150, but packaged into the length, actually smaller in length than this year's two-door Mini Cooper. If our vehicle was available today, it would actually be the smallest vehicle on U.S. roads, despite the fact that it has a five-foot bed and seats five

SPEAKER_177: people. It's even smaller than one of those tiny little two-door Chevys that I see at times? Absolutely. It's feet shorter than that. It's shorter than a two-door Mini Cooper. That's absolutely awesome. So what was the original inspiration for making

SPEAKER_180: such a small truck? And we'll show some pictures here in a second, but I'm curious why this came

SPEAKER_178: to mind for you. Well, first off, I'm a truck guy. I've driven a Toyota Tacoma 230,000 miles,

SPEAKER_179: I think, so far. I have a 180-pound dog. I live in downtown San Francisco, and I do truck stuff. I go mountain biking. I go snowboarding. I carry a lot of the stuff we use for the shop on a day-to-day basis. So having a truck, it rocks. It's actually a really useful thing to have. But I can't ever navigate downtown San Francisco. I can't park it anywhere. My wife wants to go out to dinner. It's just such a headache. It has both a financial and an emotional burden trying to navigate a downtown city.

SPEAKER_177: And we felt like we were uniquely positioned to kind of fix this. So tell me about that unique positioning. Why are you the right guy to build Tello?

SPEAKER_178: Yeah. So my background, so I grew up in the Seattle area. I built motorcycles and vehicles from scratch when I was a kid, crashed them in blazes of glory in my high school parking lot. But was studying mechanical engineering. And when I graduated, I went into automotive safety. So I worked on some of the very first autonomous driving vehicles on the sensor side, then on the software side, then on the hardware side. I ended up doing the safety systems for some of the very first electric pickup trucks that came to market. So I had a really big background in automotive safety and had a good understanding of why vehicles are designed in the way they were designed. And when you understand stuff like that, you might understand why, well, with this transition to this new energy kind of domain, we can do things uniquely capable in these vehicle platforms that have never been

SPEAKER_179: possible before. If you remove the 1,000 pound giant engine block from the front of a vehicle, repackage the front of the vehicle in a way to just support the crash safety side of things, you can rethink that entire front structural design and really shrink the footprint of the vehicle. And where that matters-

SPEAKER_42: Let's go ahead and actually, Jason, hold there, because I feel like we should just show people what we're talking about now. So if you're listening to the audio version of this,

SPEAKER_06: we're on telotrucks.com looking at their comparison tool. If you're watching the video, look at this. So here is your truck and it is superimposed right next to a Toyota. This is a Tacoma, yeah? That's right. And it's much, much, much smaller. And just for folks who are curious, this is the Tello truck in comparison to a Mini Cooper, basically the exact same length, a little bit taller, and designed to carry quite a lot of cargo. And Jason, just to be clear though, in this example, we're once again seeing a car designed here to hold an enormous engine in front, and you essentially cut off the nose and save, what's that, a couple of feet and 1,000 pounds.

SPEAKER_178: Yeah, that's exactly right. And you make it up in battery weight, of course. So it's not like we're necessarily coming in lighter weight than a Mini Cooper. But what we

SPEAKER_179: are doing is actually packaging a truck that can do truck stuff for specific areas where truck stuff

SPEAKER_06: hasn't been possible before. I think it's especially pertinent to me. We were talking before the show, but I live in Rhode Island in Providence, which city designed, I swear to you, for horses. And whenever someone drives one of these larger standard today, modern American trucks, it takes up essentially three lanes and everyone hates them. So I think this would be absolutely ideal for me. But sticking to the geographic theme, one reason why I wanted to talk to you is you guys are planning to both design and build these inside the United States. I think in a facility in Irvine, California, I, when I think about EV manufacturing, I think about a global footprint and the rise of China and all this, I was a little surprised to see your ability here. So I'm curious

SPEAKER_09: why that choice, and how hard is it to make something like this here in the States if you

SPEAKER_178: don't have Tesla scale? Well, first off, I think one of the things I understand about what Tesla came about, and my co-founder was early in the days of Tesla, both founders of Tesla invested in our company, were the only other EV company they've invested in, is they didn't start at scale. They

SPEAKER_179: started with 2,400 Roadsters that were contract manufactured from Lotus. Then they grew into 10,000 or so Model S's. So they did not come out of the gates trying to build 100,000 to a million vehicles per year. And a lot of other startups that came before us in the last 10 years felt they needed to compete at that scale and made huge financial investments without actually getting vehicles in customers' hands. And that just burned capital quicker than ever. As soon as the markets changed, they ran into huge headwinds and they were unable to actually substantiate their company. So we think it's really pertinent to look at what actually was successful in the history of automotive and say, you can't start out of the gates at high volume. What you need to start with is a product that people love or willing to pay for and find ways to get to unit profitability at moderate volumes that you can actually sustain in the way that you're actually developing. So we contract manufacture a lot of our vehicle. What we do in-house is we do all the engineering in-house. So we don't have to pay any nonrecurring engineering costs to suppliers that would otherwise charge hundreds of millions of dollars for it. We build all of our battery packs in-house, which is the number one cost driver of the EV. It's about 33% of the direct material costs of the EV go into the battery pack. So we do that in-house. We own that manufacturing in-house. And the things that automakers that have been doing for 120 years, we let them keep doing it. Stamping steel structures in the vehicle. Detroit is extremely good at that. You may remember watching 8 Mile and seeing Eminem, you know, stamping those big steel sheets and those big presses. Like that is what America's been amazing at doing in the last 120

SPEAKER_06: years, and they will continue doing it for us. That is a callback to a movie I have not seen in a while and did not see coming. But yes, Eminem does in that movie work in a Detroit stamping facility, and it looks about as interesting as you would think to have that as your career. But I'm glad that we're good at it because it makes for companies like yours possible. Okay, so this thing is going to have a 152-inch length. It's going to be able to tow, according to your website,

SPEAKER_187: 2,000 pounds, 6.6 thousand pounds. 6,600 pounds towing. Oh, I'm sorry. A payload of 2,000 pounds.

SPEAKER_06: Yes. And 6.6 thousand pounds towing. For someone like myself who has not driven a truck in a while,

SPEAKER_187: I'm not familiar with just how competitive that is. Are those big numbers or are those relatively

SPEAKER_179: small numbers? For the midsize truck market, so like the Tacomas, the Rangers, the Colorados, that would be a fairly substantial amount. Okay. It's right on par with the lower end of like the F-150s, 1500 vehicles. It certainly doesn't compete at the scale of the class 3 or class 4 trucks. But at the same time, EVs have a really interesting kind of characteristic about them from an engineering perspective where payload doesn't have a substantial impact on range. It has somewhat of an impact, but not a very substantial one. Towing, on the other hand, has a dramatic impact on range. So EV trucks, in particular, are not the best vehicles for long-haul towing when you worry about the time to get there because you have to stop for charging more frequently than you would stop to refill

SPEAKER_187: for a gas or diesel vehicle. Okay. But I'm thinking about your truck, the MT1,

SPEAKER_06: and I'm thinking about myself as someone who wants to show it to his spouse because I would like to buy one. And I don't think I would ever care about towing. Now, capacity, sure, because I might move, you know, concrete bags or maybe soil or gravel, whatever. But is towing a key use case for the MT1?

SPEAKER_42: Because to me, it feels a little bit to the side, if that makes sense.

Chamath Palihapitiya: Yeah. No, I think it's a fair assessment. I think there's applications where

SPEAKER_179: if you're using a work truck in a city and you want to go pick up a trailer from Home Depot or you want to go pull your boat out of the water, that's the use case that we feel really strongly about. That's an application we can absolutely address. We do not see this as an application where you're doing towing cross-country with our vehicle. Okay. Now, in a couple of your posts,

SPEAKER_187: you guys mentioned that fleets have shown a lot of interest in this vehicle. Do they have any towing

SPEAKER_06: requirements or are they mostly just looking for the same cargo carrying capacity that we're describing?

SPEAKER_179: Absolutely. I mean, the majority of the applications of trucks, even in downtown cities,

SPEAKER_178: are not towing applications. But there are certain instances where if you did not have it,

SPEAKER_187: it's a non-starter. Ah, so you need to have in your back pocket, even if it's not something you're going to use every day. Correct. People don't,

Chamath Palihapitiya: even commercial customers don't buy their vehicles for the 95 or 98% use case. They buy it for the 0.1%

SPEAKER_179: use case and making sure it can satisfy at least the 98% use case. Okay. That makes good sense to me.

SPEAKER_09: Now, I want to talk about cost because I went back through a lot of media coverage of the company and the only thing that I could find was an old Tuckridge article saying that before incentives, this is back in 2023, so things may have changed. You guys were thinking about 50K for this truck. Is that still the right price range for what the MT1 is going to cost? Yeah, it starts at the mid, the low 40s,

SPEAKER_179: and it works its way up there, depending on how you accessorize the vehicle, the range and the motor options. And again, the average cost of a vehicle in the US is $49,000 right now. So we want to be on

SPEAKER_178: par with what you'd expect from a cost of a vehicle. We don't think we're going to be the cheapest option available on the market. We think there'll be, I mean, we just heard the announcements that Ford's making for a new EV platform starting at 30,000. You've seen other companies like Slate

SPEAKER_179: come in with, with lower dollar amounts. We think that what we're trying to address is a specific, unique capability that doesn't exist in the market today that nobody is solving. The fact that you can't have a fully capable crew cab work truck that works for a downtown city. So that is the focus we want. And if that is, has a emotional and financial burden for you, then this is an option

SPEAKER_15: that may make a lot of sense. So is the main difference between you guys and Slate that their

SPEAKER_104: truck holds two and your truck holds four, so yours is just more capable as also a kind of a vehicle for going around the city, even if you're not using the bed? You know, I think that the

SPEAKER_179: market's really the difference between us. I think they're going after, you know, the being, trying to be a lower cost approachable entry point for a vehicle is fantastic for a collection of people that really might use that. People that want a secondary vehicle, new drivers, older people that don't have things, they get families that they're driving around very frequently. That's an excellent option for them. Our specific application is for people that need to do truck

SPEAKER_187: stuff in downtown cities. Okay. So they might have a dog they want to keep in the cab. They might have a child they want to keep in the cab. So it has to double as a car as well, effectively, because if

SPEAKER_06: you're driving around a city, you're using it for transit as well as transport. Okay, I'll take that. How big is that market? I have no idea, because I don't know how common I am, because you're really

SPEAKER_187: talking to me here, and this resonates, but I don't know if I'm one of 10 or one of 10 million.

SPEAKER_179: Well, 3 million trucks are sold in downtown cities every year. What? Yep. 3 million? 3 million. All right. I mean, we're in a pretty big market. Now, part of the challenge is when you look backwards and say, how big is that market? Well, there hasn't really, it doesn't exist anything in this market, so it's hard to say, how many people are going to buy a mini truck? Well, zero people bought a mini

SPEAKER_06: truck in the last 10 years. No, no, no. Wrong, sir. Some of my friends around town have imported Japanese K truck. Yeah, exactly.

SPEAKER_178: I should have phrased. Zero people bought in U.S. space.

SPEAKER_179: So 10,000 K trucks were imported all of last year. They're the number one most imported vehicle from Japan per the 25-year rule in the U.S., and that's actually a pretty good sign to show that there is latent demand, because if a consumer is going to go out and spend the money and effort and energy to try to import something from Japan that's not even legal, that's got 100,000 to 200,000 miles on it, it doesn't meet crash safety requirements, and that means that there's some latent demand for this product. That's actually one of the reasons

SPEAKER_06: why my thesis about your company being successful exists, because I know how many people want their little K trucks, and I know that because there's one a couple streets from me up, and I'm often walking the dogs and the kids, and you should see how people react to it. They go like this, what's that? And they get totally enraptured in the idea of having this small city car that has a

SPEAKER_09: cargo-carrying capacity. So, Jason, I agree with your thesis, and I love what you're building. Why hasn't the American major car manufacturers gotten to the same conclusion? I have a thesis

SPEAKER_219: about this, but I'm curious why you think they're not already doing what you're doing.

SPEAKER_178: Well, there's a couple of interesting just phenomenon that's occurred in the last 15 years.

SPEAKER_179: One of them is, in 2010, the Environmental Protection Agency changed their rules and regulations, where they, for the longest time, light-duty trucks were exempt from emissions regulations, and they just brought them back into the fold, but they based their emissions requirements on the size of the vehicle. Namely, the bigger the vehicle you drive, the less stringent they are on the MPG your vehicle had to get. So, a lot of automakers went, oh, man, I have to invest a billion dollars to make a new motor, or I just make my wheelbase a couple inches longer, and I meet all requirements. And so, if you look at something like the Ford Ranger over the last 15 years, it's gotten 50% larger, but 0% different in fuel economy. Well, that's just depressing, but people respond

SPEAKER_187: to incentives. Okay, so people were incentivized to make larger trucks versus smaller trucks, but even with all that, if you're making an EV, we're not talking about emission standards or

SPEAKER_177: CAFE or whatever, so why haven't they tried to do this? And that's the challenge here, is a lot of

SPEAKER_179: automakers are condensing their vehicle platforms. They're trying to build three different global vehicle platforms for every single one of their vehicle designs. So, if your truck platform is your big diesel or gas-based vehicle, well, swapping it into a battery electric platform without changing much of the other infrastructure just leaves you with that same giant platform. This would have to be a ground-up redesign, and in traditional automotive, a ground-up redesign takes at least eight

SPEAKER_191: years to accomplish. You founded the company in 2022, right? It's been three years since then.

SPEAKER_06: How have you been able to go so much faster? Just less red tape, less historical baggage, or did you actually change some fundamental thinking to get this truck to where it is today so quickly?

SPEAKER_178: So, there's a term that was coined by

SPEAKER_179: an automaker. I won't call them that by name, but that was virtual validation, meaning that they were going to build and validate their vehicles entirely in software before they built any hardware. They were going to use the state-of-the-art tools to do that. But when you have an automaker that employs a hundred thousand validation engineers, trying to say, hey, we're going to remove all of you validation engineers. We're going to move all of the software. It's just an impossible thing to do.

SPEAKER_178: So, the fact that automotive design cycles take three to five years and then the engineering cycles take another three to five years, now you've got an eight-year product cycle from inception to

SPEAKER_179: actually deliveries, it's largely because of the way that we've moved from hardware design, hardware validation, and implementation of manufacturing protocols. What we're doing is we're building the entire company from the ground up using software to build and validate our company,

SPEAKER_178: and also to use some of the cream of the crop and AI tools. For example, crash testing just 20 years ago used to be build something, a portion of the vehicle or a scale model vehicle, slam it across

SPEAKER_179: as many times as you can. Every time you have to rebuild something, that's a ton of engineering design cycles to rebuild that and just slam it into the... That's why things take so long. Well, recently we've gotten so good at physics-based simulations that we can now crash a vehicle, but that still takes like 10,000 compute nodes and 24 to 48 hours to compute. And so even those at scale still don't... They're much faster than the months-long process, but they're days-long process. So the question we've posed is how do you make that a minutes-long process? And that's a lot of how we implement how we do all of our build and design and validation.

SPEAKER_224: Okay. And by the way, it was Porsche who said virtual validation, right? It was not Porsche, but I won't play it. Wow. Well, it proves what I can Google while I'm also paying attention at the same time.

SPEAKER_06: Okay. So we talked about scale earlier and how you're not going to go after the day one mass manufacturing approach. You're going to start smaller and then build from there. I know you guys said in 2023, you had 500 pre-orders that are scaled to, I think, 2,000 in 2024. How many pre-orders are you guys at today? And what does a first production run look like in terms

SPEAKER_178: of scale for Tele? I think we're just under 12,000 pre-orders right now. Okay. And so our goal has been to get our first vehicle in a customer's hands in 2026. That has always been our kind of milestone. It's probably looking towards the end of that year that we're going to

SPEAKER_179: do those deliveries. We will deliver a small batch of vehicles to first early access customers that will probably be somewhat incomplete from a software perspective. So these won't be a mass run of vehicles. We'll have our engineers stationed with them and making sure that those get to a state where we're very happy with. We'll then build 500 right after that. But if we feel happy with that, then we'll build 5,000. And 5,000 is really the break even point for the company where if we can build 5,000, we can be unit profitable on the vehicles.

SPEAKER_230: And then just because I'm an accounting dork, how many vehicles would you need to have enough gross

SPEAKER_09: margin on them to actually pay for the operating side of the business as well? Is that 10,000 or is it something more like 50?

SPEAKER_179: Yeah. So I've financially modeled this to the nth degree, but it's between 10 and 20,000 to actually get corporate profitability. But there's a lot of options open to us once we hit those numbers. We will probably start bringing more and more in-house after we hit unit profitability, which has its own set of operating costs that are associated with the work we're doing. So it's not

SPEAKER_187: going to all happen at once. But once you bring stuff in-house, you don't pay someone else's gross

SPEAKER_09: margin and you can over time lower your bill of materials, labor costs, and get more efficient.

SPEAKER_179: Yeah? Of course. But there's a capital expenditure with doing each of those things. So it would have to coincide with the financing as well.

SPEAKER_06: Which is actually one thing I wanted to talk about. So whenever I talk to a founder, especially if it's a company that I haven't talked to before, I always go through their fundraising history to figure out who's backing them, how much are they raised, gives the idea for burn and so forth. I was only able to find a couple of relatively small funding rounds for Tello. Yeah. And either I'm missing some numbers and you guys have raised money I don't know about, or you're the single most capital efficient company of all time. So how much money have you guys raised

SPEAKER_187: so far? And how much more are you going to need to get to that 5,000 vehicle break-even economics point?

SPEAKER_185: So both the things you said are true, by the way.

SPEAKER_179: Oh, okay. There we go. With spending only $6 million in the last 18 or so months, we've built two on-road vehicles that are registered through the state of California. We've built a battery MPI line. We've engineered an entire set of vehicles for a beta and gamma build of our vehicles. That said, we have not, and I unfortunately can't tell you too much until a few weeks from now, but we've definitely been in the finance, the fundraising process for quite some time.

SPEAKER_06: Okay. You can't get specific, but you could probably give me a guidepost here. Do you need to raise tens or hundreds of millions of dollars? And I'm asking because I saw Rivian go through its life, go through its IPO, and I've been tracking its, I'll call it impressive capital consumption as it

SPEAKER_178: works towards functional scale. So the fact that you even asked it in the tens of hundreds is actually,

SPEAKER_179: I appreciate that because that is how we're thinking about those numbers.

SPEAKER_178: It's in the tens. It's like, we will need to raise before unit profitability at least another

SPEAKER_06: hundred million dollars. That's not that much for what you're doing, because car companies scale in revenue terms very quickly because you're selling $50,000, just big round number, $50,000 units. So the revenue scale is pretty neatly along with that. That still feels cheap, frankly.

SPEAKER_177: Yeah. Huh.

SPEAKER_179: Again, to be clear, we had 11 people in the company up until recently, only 11, and that got us to where we are today.

SPEAKER_06: All right. Well, if people want to learn more, it's telotrucks.com, T-E-L-O-Trucks.com,

SPEAKER_09: and I think they're just fantastic. Jason, all the best. And when you start manufacturing, I'd love to have you back on so we can see what the factory floor looks like.

SPEAKER_06: Yeah. Wonderful. Awesome. Thank you.