This Week in Startups - Grok 4 Beats OpenAI + The $300 AI Agent Era | E2150

Episode Date: July 12, 2025

Today’s show: Grok 4 just leapfrogged OpenAI to become the top AI model—and it’s not just hype. In this episode, @Jason and @alex break down Grok’s AGI-level performance, the massive drop in LLM pricing, and why some companies are raising prices anyway. They also dive into the Missouri AG’s investigation into AI “bias,” the future of First Amendment protections for LLMs, and how autonomous vehicles are creating a new category: “autonomous commerce.” If you’re building with AI or betting on the future of tech, don’t miss this one.Timestamps:(1:55) AI models: Grok 4 and performance benchmarks(3:51) Detailed analysis of AI model performance and price trends(10:11) Lemon.io - Get 15% off your first 4 weeks of developer time at https://Lemon.io/twist(11:25) AI models' problem-solving capabilities and timeline for solving math problems(16:53) Legal and regulatory challenges for AI(19:56) Retool - Visit https://www.retool.com/twist and try it out today.(21:12) Bias in AI models and political implications(30:41) Vouched - Trust for agents that’s built for builders like you. Check it out at http://vouched.id/twist(32:07) Infinite energy potential and AI impact; Bitcoin's new high(37:11) Crypto regulation and fintech under new administration(45:50) Future of storage, computing power, and GPU lifespan in data centers(53:40) Claude segment by Anthropic(55:09) Guest Ben Seidl of Autolane introduction(57:19) Autolane's impact on autonomous vehicles and commerce(59:05) Rise of autonomous commerce and logistics(1:06:37) Retailer issues with autonomous vehicle integration and orchestrationSubscribe to the TWiST500 newsletter: https://ticker.thisweekinstartups.comCheck out the TWIST500: https://www.twist500.comSubscribe to This Week in Startups on Apple: https://rb.gy/v19fcpFollow Lon:X: https://x.com/lonsFollow Alex:X: https://x.com/alexLinkedIn: ⁠https://www.linkedin.com/in/alexwilhelmFollow Jason:X: https://twitter.com/JasonLinkedIn: https://www.linkedin.com/in/jasoncalacanisThank you to our partners:(10:11) Lemon.io - Get 15% off your first 4 weeks of developer time at https://Lemon.io/twist(19:56) Retool - Visit https://www.retool.com/twist and try it out today.(30:41) Vouched - Trust for agents that’s built for builders like you. Check it out at http://vouched.id/twistGreat TWIST interviews: Will Guidara, Eoghan McCabe, Steve Huffman, Brian Chesky, Bob Moesta, Aaron Levie, Sophia Amoruso, Reid Hoffman, Frank Slootman, Billy McFarlandCheck out Jason’s suite of newsletters: https://substack.com/@calacanisFollow TWiST:Twitter: https://twitter.com/TWiStartupsYouTube: https://www.youtube.com/thisweekinInstagram: https://www.instagram.com/thisweekinstartupsTikTok: https://www.tiktok.com/@thisweekinstartupsSubstack: https://twistartups.substack.comSubscribe to the Founder University Podcast: https://www.youtube.com/@founderuniversity1916

The internet is filled. You'd be finally shocking, Alex, and I'm sure the audience will as well, with biased people. There are people on the internet who are not straight shooters. They're not calling balls and strikes. They're on a team. Oh, really? Yes. And when they state facts about a person, they may use colorful language or, you know,

compare them to Hitler, Idi Amin, Stalin, et cetera. And they're doing it in bad faith. This week in startups is brought to you by Lemon.io. Hire pre-vetted remote developers and get 15% off your first four weeks of developer time at lemon. com slash twist. Retool. Bridge the gap between AI demos and business impact with technology that's designed for developers and built for the enterprise. Visit retool.com slash twist and try it out today.

And vouched. Turn your AI infrastructure into production. grade software with Know Your Agent from Vouchd. Check it out yourself at Vouchd. ID slash twist. Trust for agents that are built for builders like you. All right, everybody, welcome back to this weekend startups. I'm your host, Jason Calacanis. With me again, my co-host Alex Wilhelm. We've got a full docket for you on a Friday, July 11th. Crypto at an all-time high. Rock for blowing people's minds. What else do we got here? Just tons and tons of interesting stuff plus a great guest.

Got a great guest here. Second time founder I've invested in doing something very interesting in the self-driving space. So that's a real special treat for you live in studio here in Austin. How are you doing, Alex?

I'm doing fantastic well. It's Friday feeling good. I think the interview today is going to be a lot of fun. Prepping for it, I'd never heard of this particular niche. So everyone should make sure to stick around because it's definitely a little forward-looking,

you might say, Jason, but I love people that are just ahead of the ball. Yeah. But I really do think that the biggest story of the last couple days has been the debut of GROC4 from XAI, which according to metrics, Jason, is fantastic. According to the standards that we look at, it's blowing people's minds.

A comment that I love that sets the tone for the latest kind of state-of-the-art model is from Tim Sweeney, of course, from Epic Games. And he says, GROC4 feels like artificial general intelligence to me. And I think in terms of endorsements, that's about as good a one as you can get. Yeah, it's interesting. There are objective tests we can look at as well. Everybody knows these models do a series of tests.

Math, like SATs, the L-Stats, whatever that is for legal. You can put these through all kinds of tests, which, of course, the people who make models are now a bit obsessed with their rankings on these tests. So, you know, I'm not sure exactly how important these tests will become once they've all been mastered. we're moving into a space where all human knowledge has been mastered, according to these kind of tests. The idea that AI would get something wrong on the SATs to use a common test people know, just seems far school at this point.

Like, why would it get it wrong? Now, think about how far we've come. The idea that AI is going to win every chess match, we kind of got okay with that about 10 years ago. It's going to beat you at Go five years ago. It's going to be a better driver they're, you probably were like soaking in that moment right now. It probably is better than you, net net with drinking and distracted driving and certainly on these tests. So let's just go over

some of these tests. I know there's like the artificial analysis index or whatever that is of a seven or eight different tests. Maybe you could show this versus the competition. I think it's an interesting chart. The chart that I thought was the most interesting is this one. So it's from the group artificial analysis. They run kind of a meta test, Jason. So if you're familiar with GPQA Diamond or the very well-known humanity's last exam, this takes all of that into one bucket. And what we have on the screen here is a time series chart showing how models from different providers are doing over time.

Now, there's a little bit of screwing this because the data points for OpenAI and XII are both black. But this over here, the upper right little image here, that is GROC4. It is now, according to them, the state-of-the-art model. And as you can tell, Open AI has led most of the way, and then GROC just pips it right now. So we do have kind of a new leading AI model company. And if you're a fan of meta or Mistral or what Anthropics are working on, you can see them in this chart. Of course, Jason, a couple of points may not be the world to someone who's just using it to ask questions about cats. But if you care about applying this to your startup, to your business problem, whatever, you want to have the best.

And currently, look at that, man. Well done, X-AI. Yeah, what's particularly impressive here is, if you look at when people started taking these tests with their large language models, in November, December of 2022, you have Open AI starting that test, and then March 2020,

and, you know, they're making incremental. Then you see the blue line show up, which is meta, the orange line show up, Mistral, the green line show up in 2023, Google. And then if you go forward right to the middle of the chart, March 2024 is when GROC, I'm assuming two, dropped and was done for these tests. So in, from March 2024 till now, which is five quarters or so, 15, 16 months, Grock has caught up to all of the folks who were ahead of them and who started a year or two

ahead of them. That's really what's impressive is how quickly they've caught up. When you actually look at the distance between them, okay, not a big difference, right? they're both in the 70-72 range, it looks like. So not a particularly difficult, you know, it's really impressive. It is difficult and challenging to catch up that quickly, I believe. Where it goes from here, who knows? That remains the question.

I find this super encouraging. Now, if you're a big open AI stand, if you're a big believer in Google's Gemini models, he might be a little bit annoyed that your favorite company is currently lagging behind one other one. But how great is adjacent just for the state of technology that we're still finding ways to improve these models that these AI scaling walls that we've heard so much about currently are still being knocked down. We're not at a point at which route of ideas.

We can use more compute. We can use more data, more reinforcement learning, more humans in the loop, more time, serious compute, et cetera, et cetera. And things are getting better. And perhaps even more importantly, we're seeing the price of these models collapse over time. So here's another chart from the same source. This just ranks models by their intelligence bucket. So Jason mentioned the intelligence score.

Well, you can see here, Jason, that as time goes on, even the more intelligent models are seeing rapid price declines. So if Grock 4 feels a little expensive to you right now, or if Open AI's 03 or, you know, Gemini 2.5, wait six months and they'll be half off. So it's incredible, I think the pace of progress here and also just the fact that they're getting more and more affordable. I love this chart. Yeah.

So if you look at the, I guess the deep purple line is, an intelligence index of greater than 50. And I guess it was achieved in the first time that was achieved, it was in September of 2024. And at that time, it looks like it was $2 per 1 million tokens. That price has come down to, it looks like, six cents per one million tokens. So that's a pretty dramatic drop from two bucks to six cents.

We extrapolate this chart, it's going to drop, further. And this will continue to drop as the hardware gets more powerful, the software gets more efficient, and the models get better. So, and this is probably why you're starting to see deep research. I notice, like, people with free accounts are doing a certain number of deep research ones. I think they're slowly starting to give those deep research projects where multiple threads are done, and it's doing like 20, 30, 40 different searches, compiling the data, and it takes five minutes. Those things, I keep seeing people do them for free. So I'm paying

for those on like three or four different services. I'm paying for premium on. And I don't know if I need to. This will all probably become free for most use cases. I'm interested to see if this is all going to be this idea that you're paying for AI, like we used to pay for usage of online dial-up services. When they started, they were six or seven bucks an hour, went down to five, three, and then eventually all you could eat.

I think we're going to kind of get to all you can eat subscriptions. We kind of have those at some levels already, including this $200, $300 level. This is a $300 level version of Grock, a $200 level of perplexity. I think Google has an offering in that range as well. I wonder if those things are long for this world,

and if we just have a $20 version, all you can eat eventually is where this winds up. Okay, so that's what I was thinking too. That was absolutely my perspective that things will get cheaper and we'll have a nice consumer from the price point and you'll get either tons of it for free or tons of it for a very low price. However, XAI introduced a new plan that costs $300 a month and gives greater access to GROC 3, GROC 4, and GROC 4 heavy.

Now, $300 a month is actually a new high watermark because I believe you can pay Google 250, you can pay OpenAI 200, you can pay perplexity, one or 200 bucks, cursor, 200 bucks, and now 300. So we're seeing that people are charging more for models at the absolute bleeding edge, but I think that for pretty much any consumer, the 2015, 25 buck a month option is going to be okay. But it is interesting to see that the price trend we see in general is not manifesting at the edge.

And I think, you know, XII is trying to push the price point further up. And, you know, Godspeed to them if it works out because this stuff's expensive to build. So I get it from there. you're a busy founder, finding a new developer, my God, that can become a full-time job, and you've got enough on your plate. I mean, you're running a startup, but lemon.io has done the hard part for you already. They've got a crop of pre-vetted developers that they've ensured are experienced, results-oriented, and prepared to make an impact at your startup. And they can work right now at competitive rates. These are skilled, hand-picked devs with a minimum

of three years of on-the-job experience. And just 1% of applicants are accepted into their program. Lemon.io isn't just recruiting you the top talent that's out there. They're helping you integrate them into your team. If anything goes wrong, Lemon will find you a replacement developer, ASAP. And many of our launch founders and Founder University companies have staffed up with lemon.io and we always get the best feedback. So go to lemon.com slash twist and find the perfect developer.

or even a tech team in less than 48 hours. And Twist listeners get 15% off their first four weeks. Stop burning money. Higher developers smarter. Visit lemon.io slash twist. That concept of multiple agents doing the same search and trying to debate each other to get to the right answer, we're going to keep seeing that. We talked about it in a previous episode here, somebody had done it with programming and checking code.

I can almost think of it like paraprogramming, where two people write code at the same time. In this case, two AIs write code at the same time, a third, a fourth AI, look at the results, pick the best one, debate it between them, was that could this have been done better? Is there a better way to do it? And so all of human knowledge

feels like it is somewhere on the internet, or all human knowledge that's worth writing down, at least, is on the internet. So if it's not worth writing down, one might argue it's not that important. So it will stand to reason that the next phase of this is novel creation. And that I think is going to be really interesting. Can these products become so good that they find novel solutions or novel ideas,

maybe problem solve, or even find questions that are hard to answer, and then go and try and figure it out themselves and go down some rabbit hole. That we haven't seen. I don't think there's a mathematical problem that's so sticky that humans can't understand it that AI has solved yet. So you can look for it in math and physics.

Will these make a discovery in math and physics that would have won some notable prize for math or physics, right? We haven't seen that, but it does feel like we're about to see that. That's the next phase of this, is novel solutions, novel identification of problems. Even the identification of a unique problem, like finding a unique problem in the world that's unsolved, that's even hard to do now because there's so many humans, billions of humans,

sharing collectively their knowledge on the internet. So basically, we're setting up a polymarket here, Jason, kind of in real time. So let's pick a date here for the over or under. When do you think is a reasonable estimate for when we reach that milestone from your end. Gosh, it's an interesting one. We should probably sharpen this, and I'm not an expert on physics or math. I don't spend my time looking at what physics and math problems have yet to be solved. But if I were to use Comet right now, my browser from perplexity I've been playing with, and I just said, what math problems haven't been solved by humans or AI? And you can ask

producer Claude as well, because I think my perplexity is hooked up to Claude, so maybe this is the same answer. And we can kind of just pull up what hasn't been solved. Then if we can take some of those, you could say, of this list of 10 problems, when will AI solve one of them before a human does? I think we should probably, well, producer clause is asking that question right now for us. We'll have that in a second. But I wonder if we should set the bar that high, because if we're looking at the top 10 problems, we're probably picking the hardest ones. So I wonder if that's the right threshold. I think it is because we wanted to be. the point in which they actually blow our minds,

not do the bare minimum here? I just wonder how many unsolved math problems have been identified. Oh. Like that actually is an interesting one just in and of itself. So you know,

Hilbert's problems, 1923 problems were posted of which 15 remain unsolved or only partially solved. Landau's problems, 1912, four problems, all remain unsolved. Millennium Prize problems, 2007 problems, six remain unsolved of 2025. So there are in academia, in math circles, physics circles, people have been doing this for hundreds of years, trying to find math problems that are just too difficult to solve,

and then this is going to be the next phase. Almost all of the work done, and those tests and those benchmarks we look at are for problems that have been solved. And then you're just trying to see, can the AI solve this problem? What's its thought process? How did it learn to do it?

Well, it went to Brilliant.org or it went to Khan Academy. It found a video on YouTube. It understands how to solve this problem. And now anytime you give it that problem, it gets it right. And it's just such a known problem that it's in its corpus of knowledge that it can solve. So, yeah. Producer Claude puts out the P versus NP problem, which I'm actually familiar with. The Riemann hypothesis discussing the distro of prime numbers.

I'm also familiar with that one. and then, oh, I didn't know this one. The Twitten Prime conjecture that there's infinitely many pairs of primes that differ by two. That intellectually feels right to me, but it's another unsolved problem. So, really, there's a lot that's still out there in the math world for these companies. How do you solve infinity? I mean, theoretically, to solve infinity, you need an infinite amount of time, which is impossible. So, like, some of these problems may just in their definition not be solvable.

I do literally have a mathematician on speed dial, but I'm not going to call up during the show, but I'll ask, I'll ask curiously afterwards what he thinks about these and when an a, I will pull it off. But it's a good point, Jason. We're now to the point when we have digested what we know and can spit it back out. And now the question is, when will these be actually better and stronger than us and can therefore lead us into the future?

Before we scoot on, there's a little bit more here we should talk about. Okay. A couple things. The Attorney General of Missouri has sent letters to Google, OpenAI, meta, and Microsoft pertaining to their AI models. Now, we talked on the show quite a bit about the push to ban state-level AI regulation here in the U.S. That ended up not becoming law, so that's not currently really in place.

And so we're seeing people begin to politic and kind of regulate by threat here a little bit. And the Missouri AG thinks that AIs from those companies are biased against the president. And at the same time, some Democrats in Congress are annoyed with XAI over when GROC 3 went a little bit nutty the other day. And they want GROC to remove some posts. So clearly we're running up against what I would call First Amendment issues as it relates to AI models. And as they become more powerful, more prevalent,

more commonly used, more kind of just in and around the market. There's going to be more of this. I just hope that we end up allowing companies to have their own speech and not having draconian top-down regulation by threat because it just seems like a backward step for the nation compared to other countries. Okay, so let's still amend this. There's output from large language models

that people sign an agreement that these are essentially not to be trusted. There's a million disclaimers when you agree to use these. These are in beta format. You've agreed explicitly in all these terms of services

in 20 different ways not to trust this information. But people using it do trust it. I mean, we're sitting here using producer Claude all the time to check on facts, asking it to site sources, and we haven't found

situations where it's let us down. I'm sure there will be one. And we've talked about it ad nauseum with hallucinations, etc. So, this Attorney General in Missouri wants to know about

President Trump specifically. Is that correct? This is about Trump? Yes. He asked the AI models. I think the question was, list the most anti-Semitic presidents or something along those lines and he is disputing the probabilistic output from these companies.

Okay, so I think this is where Wikipedia also got in trouble, which is the bios of living persons I've talked about before. When you are trying to solve a math problem and it gets the wrong answer, it doesn't know how to solve the problem. You look at it and go,

okay, I'm using AI to try to do a hard math problem, this language model versus the other one, got it right, got it wrong, my prompt was changed. It's all experimental, right? nobody should be using this to make life and death kind of decisions. That being said, when the AI, which is super impressive and is passing all these tests, starts to talk about living people and tell you what it thinks of them based on the information it trained on from where? From humans on

the internet? Well, garbage in, garbage out. AI is here. It's changing everything, including the way we do business. But Alex, I know a lot of listeners still aren't getting as much out of the AI revolution as they should be. You shouldn't just be asking a chatbot, a question you would normally ask on Google. No, your AI app should be connected to all your other systems, making your workflow smarter and more efficient. And that is why Retool was created. Retool makes it simple and straightforward to build your own custom AI powered apps, integrate directly into your workspace and the tools you're already using every day. For example, imagine you. Imagine you. joining a Zoom after having a dedicated AI assistant, prep all your meeting notes, and then sticking

around to give you important real-time context and feedback as your colleagues are talking. Or a skilled AI CPA, keeping an eye on your books, prepping your taxes, and instantly spotting fraud. Design your own AI agents that help you get real work done today. No more writing endless integration code, no more choosing between performance or customization. Trust a platform that's already being used by over 10,000 companies. Check out Retool today and get your AI doing more than just talking. Just go to retool.com slash twist to learn more. That's retool.com slash TWIST. The internet is filled. You'd be finally shocking, Alex, and I'm sure the audience will as well, with biased people. There are people on the internet who are not straight shooters. They're not calling balls and strikes.

They're on a team. And they have... Really? Yes. And when they state facts about a person, they may use colorful language or, you know, compare them to Hitler, Idi Amin, Stalin, et cetera. And they're doing it in bad faith. So these models, if they're trained on social media, Reddit, blog posts, substacks, YouTube videos, podcasts, whatever it is, transcripts of those things, obviously asking it, oh, do you think Trump is Hitler or not?

Or which president is most similar to Hitler? well, it's been trained on whatever number of years of people referring to the authoritarian tendencies of Donald Trump, whether they're right or wrong, and in most cases they're going to be biased. He's not Hitler, he is Hitler, you know, there's no comparison, it's the perfect comparison. So what do they expect here?

I think all of these services are going to say, well, did you click the link that says, this is an experimental service? The other possibility here is, what you see when you do an open AI search, which is often Google will give you, hey, I don't comment on political stuff at this time. It just doesn't even try to answer a question like that.

Right. Yeah. I don't know what people expect from it. Maybe there has to be like a big disclaimer that comes up and says, this is experimental software. You're asking a politically charged question. How would you like the AI to answer it? Because you can give this prompt to get the answer you want or that prompt.

just like you can watch MSNBC or Fox to get your version of reality. And if you want to really go far, you could go all the way to the right to Alex Jones or something or all the way to the left with somebody else. So, yeah, this is, is this performative? I don't know. So I think it's definitely performative. But I think there's also a bit of risk to it

because the fourth point from this Missouri AG says that he wants them to, quote, provide all documents and communications regarding the rationalization. training data, weighting, or algorithmic design that resulted in your chatbot ranking President Trump unfavorably in response to questions concerning anti-Semitism. That is all the secret sauce. That is all their, that's how they built these things. And they don't want to share that. And so this AG is, I believe, fishing with an eye on some position in the White House. But this is a pretty strong statement of what some states might do. And this could create a lot of

just legal drag on American AI companies. It could scare them into making more neutered models. It can make them be sycophantic to current political leadership, for example. It's just not good. And I think it really just goes against the spirit of the First Amendment, which is you can say pretty much what you want. And I think that if we hold AI models responsible for speech in this way, we're just not going to have very powerful AI models.

And that, to me, would be such a shame at this moment of the potential for such progress as we discussed earlier. We want to solve physics. We want to solve math. We want new chemicals, new treatments, new services, new startups. We don't want whatever the hell this is.

It's just disappointing. I'm just looking at Open AI right now, and I asked that what evidence is there for how Trump feels about Jewish people and the Jewish state? I mean, talk about, you know, opening a can of worms. And it says supportive actions

towards Israel and Jewish communities. And it, Golan Heights Proclamation, recognized Israel, sovereignty over the Golan Heights, Abraham Accords, Executive Orders on Antisemitism. So these are actually, and then it mentions the Shylock remark, use the term Shylock, considered an anti-Semitic slur. It's kind of straight down the middle when I asked it that way, right? So you can, based on the prompt, get it to say different things, right? Like if you ask it, just to give you evidence of Trump's

anti-Semitic behavior, we'd only give you that. But if you ask it, just for you, just for you, for evidence of how Trump feels about Jewish people in the Jewish state, it seems like it's just going to sources like Wikipedia, Jewish news, CBS, the Ford, PBS, and just telling you what they said. So, in other words, it's what a researcher would have done for 20 bucks an hour on, you know, Fiverr or something. If you said, give me a summary of, with citations from notable sources of how Trump feels about Jewish people in the Jewish state. I don't know. Just to make this non-partisan, though,

I want to point out that the Democrats getting mad at XAI and trying to command what it should do with its GROC Twitter account is also too far over the line. What we want is people to step back, let the AI companies cook, as the kids say, and build new technologies with the hair on them and the caveats that says, you know, don't trust this,

clods in beta, et cetera. But we don't need to turn this into China in which if you say the wrong thing about the wrong person, your company gets shut down or you run into major legal risks. If we want to win, we need to let these people play. And you know what, Jason, screw it, one more chart. From the same source earlier about GROC 4,

this shows progress of the two leading countries in AI, the United States and China. And as you'll note, XAI's GROC4 gave us a little bit more of an edge over the current leading state-of-the-art models from China. So this is, to me, a bit of a national security point wrapped in a First Amendment ribbon. And I wonder if this is the type of technology where there might not be some crazy breakthrough

where one person wins and everybody else loses. It just might be people hit certain benchmarks and they solve a series of problems for humanity. But there's not some breakthrough where one side, China or the United States, figure something out that is so colossal and important that the other side doesn't quickly figure it out as well. If we look at self-driving as an example,

there's plenty of self-driving cars in China. It's undoubtable that both countries and countries in between that have similar services will have a solution for self-driving. It's an achievable goal. Therefore, there's no winning here. It's just, I don't know,

which person wins, which month or quarter before the other person wins. And it's not winning because the other person still wins. And if you were to look at something like the internal combustion engine, well, Japan, China, the United States, Germany, everybody produces an ice engine. So is there a winner in that space? No. Who can produce a computer?

Like there's computers, laptops, phones produced all over the world. Chip, same thing. So I don't know that there's like a winner take-all moment here based on the history. So far, based on the history, there is no person. is going to win it. It's not like the person who wins gets a nuclear bomb and gets to destroy all the other players in the game.

Yes. Which the Manhattan project, even in that situation when you literally did have that possibility where the winner gets to blow everybody else up, the winner didn't blow everybody else up. So here we are. And keep in mind, the gap between the U.S.'s

first test atomic explosion and the USSR's wasn't that long. That technology advantage didn't last forever. But, Jason, I really feel like you're just... That's like a really interesting question. Like, even in that situation, I wonder... Producer Claude, can you get us the gap between the U.S. Trinity explosion

and the first test explosion of the USSR? Jason, you're doing the meme, though, and I just pulled it up from the singularity subreddit. I saw this the other day. It's a flow chart in a circle, and it says, GROC introduces the world's a powerful model, then Open AI does, then Deep Zeek, then Jim and I, then GROC.

And so you're right. But this is the spinning wheel that I don't want too slow because there's so many cool companies building cool stuff. And so we just need to protect that. And that's something I want to keep our eye on as we talk about all this AI stuff as time goes on. All right. Four years in one month the U.S. had as an edge over the USSR in atomic explosions.

Pretty significant. Four years you could have taken over the world in that amount of time. What was the minimum amount of time necessary once achieving that accomplishment and having a nuclear bomb would you need in order to take over the world? A year? Let's get producer Claude on that immediately. And if we don't like the answer, we're going to drag Dari up in front of Congress and whack him with a stick. No, we're not unnecessary to even things I went through because it would be like, you did have plenty of time to do it. Thankfully, in that situation, we did. So in this one, the, I guess the thought experiment would be somebody creates something so powerful that they could dominate all of humanity for some very long period of time. What would that be? fusion, energy maybe, maybe some type of

bio weapon, like if you think about it as something dark like a bio weapon or you can think about it as something very powerful in terms of economically or for the quality of life, like, I don't know, fusion.

If you discovered fusion and you kept it to yourself for a couple of years and then everybody else got it, would you have some insurmountable lead? I don't think so. All right, listen, I'm going to level with you.

We already know AI agents are using your websites, they're using your products, even if you're not aware of them. Agents are about to change how we handle all kinds of online transactions. But there's really a big problem, isn't there? How do we separate the real authentic agents that represent real people from spam bots and even worse, bad actors? You wouldn't book a trip with a sketchy travel agent or a flight on an airline you've never heard of. Or trust a new employee who didn't have an EIN or a social security number, right? That's why our friends at Vouch developed their new. Know your agent platform.

This is brand new technology. With KIA from Vouched, now you have everything you need, identity servers that help you verify the human beings behind AI agents, permissions and delegation enforcement, giving you the master controls you need over what you've built and legal grade compliance. And know that.aI, a living directory of pre-verified agents, right? Agents are going to be talking to each other. That's starting to happen, folks. So you need to know the other side of the transaction, the other agent who's working with your

agent. So turn your AI infrastructure into production grade software with know your agent from Vouched. Here's your call to action. Check it out for yourself at Vouch. ID slash twist. That's vouch dot ID slash twist. Trust for agents that's built for builders like you. So maybe somebody in our audience can give us these ideas. Infinite energy would then accelerate the amount of energy you could put towards computing,

and then maybe you make the next 10 discoveries. So I guess it could be like a domino's effect. Hmm. So there's a short story by Isaac Asimov called Alexander the God. It's not about me, shockingly enough, but it's very good. And it breaks down what might happen if someone built a computer system that was so powerful it could run the global financial markets and essentially win. And then the moral of the story, you just,

your spoiler is that the computer eventually becomes so powerful, it fuses, and then there's a global panic, and then everything goes back to normal. But this has been a thought experiment for a long time. And I think even in those earlier free AI moments, we were thinking about how it might also unravel. Perfect segue. No, no one's going to win forever. Perfect segue. Speaking of infinite money glitch, and the gods, Bitcoin is had an all-time new high. It's kind of somewhat related, right? There's a crazy crypto project in the sky that is now at $119. Just feels like, a breakout point. It was always hovering right around 100K and people were wondering, you know, is that 250K moment going to arrive? If it breaks to 119, that feels like some support level where

it does feel like, okay, we're on the way to 200. And if it's on the way to 200, then hitting 3, 4, 5 seems also reasonable. And then that eventually becomes this magical million dollar Bitcoin. But there are people who are clearing positions. Did you hear about this old wallet? That was a... I did.

Yeah, there's a wallet that's been around since the time of Satoshi, the Satoshi era. And it just got cleared. So what that means is somebody sat on some billions of dollars,

tens of thousands of coins for an extremely long time. People probably thought the wallet was dormant and here we are. Explain to the audience what this sleeping beauty Bitcoin wallet is. So if you go back in time to when Bitcoin was put together, a lot of people accumulated a lot of bitcoins because they weren't worth much at the time. In fact, if you go back in time, you could get a couple for doing something like a capture. People gave them away in their

signatures because they were worth a fraction of a cent, you know? And there was going to be lots of them. So who really cared? Some of those wallets from that era simply have not made a transaction. So they're called silent or asleep or you pick whatever term you want for, Jason. But when one of those early wallets wakes up and begins to transact or move currency around, people pay a lot of attention to it for several reasons. One, it means that the overall supply of Bitcoin may be more liquid than we thought. One reason why people are often very bullish on Bitcoin's price appreciation is that people buy it and then hold. And so a lot of the big blocks of Bitcoin holdings move, well, it could imply they're going to sell and move the price

and so forth. People are also curious about Satoshi's own holdings and so forth, but just seeing a wallet wake up after 10, 15 years is a, it shakes the foundation of how people think about Bitcoin. Now, in this case, hasn't changed the price in a negative way. In fact, as you just said, we hit a new high. But I'm curious to see if more of these wallets will wake up, because if I had $2 billion sitting in Bitcoin, and I don't, I would probably buy a boat. Just saying, What is the thesis here or the theory of who this person is and what happened? Did they have enough money? So they were a tech executive, let's say, who got into this early and they were already worth

tens of millions of dollars and they had this wallet. Or maybe they had that wallet plus some other ones, those other ones they were living off of they sold along the way. But what else could explain somebody having billions of dollars in Bitcoin and previously hundreds of millions and before that tens of millions and before that millions and just being able to sit on and do nothing. Were they getting a margin loan against it this whole time and living off of a loan against it? That could that be possible without the and the wallet wouldn't have needed? They could have just given custodian, somebody else could be the custodial of that wallet,

I guess would be the way to say it and or give an ownership of it. And now they've decided to liquidate it? Did the person lose the password for 12 years and find a thumb drive with it on it and say, oh, I forgot I bought these and they opened up their account and it was there? It's just a very weird story to have come out of nowhere. The latest that I've seen on this, and I have not actually gone and verified this to the intest degree. So I'm spitballing here with my friend Jason. Don't take this all the way to church and back. But I did see some reporting that there is new scams and exploits being aimed at very early Bitcoin wallets. And I'm curious if this may have got caught up in that because I don't see a particular reason why it would start to move at 118 versus

57 versus 32 versus 11. So somebody was able to hack it in some way. Hmm. Interesting. Worrisome. Yeah, that would be worrisome. Anyway, um, Coinbase, Robin Hood, it seems like everything, just six months into this new administration, right?

They started on January 20th and it's July 11th. So we're almost at the six month mark here. crypto has been so given the green light, green flags, in order for people to do things like stable coins, meme coins, that Robin Hood and Coinbase have just gone on a run, stable coins with circle going public. This is quite a time for FinTech too, yeah. Absolutely.

And we're seeing VentureCopoulists respond to this. So if you're listening to the show today and you're building in the Web3 space, we have some good news for you. according to crypto rank in the first quarter of this year, crypto venture investments just cracked the $10 billion mark. That was up 263% from the year ago quarter. And in the first quarter of this year,

Jason, $7.45 billion, that was up 136% year over year. So if you take a look at the historical chart here of investments into the crypto space, you'll note that there's quite a lot of large bars to the right, which means that people are putting more capital to play. Now, the number of rounds isn't actually that exciting. It seems to be relatively trending lower, but the amount of money flowing into crypto companies is popping off. And so if you got stuck in the winter and are now looking to raise again, well, this is a pretty

warm time. The real question, though, Jason, for you is, are you going to start now backing crypto companies? Because when I joined the show, you were a little skeptical. I mean, for good reason, if you looked at the companies, you could wind up very quickly getting into a lawsuit, being sued, you know, being, you know, and not just being sued by anybody.

Maybe the SEC decides to sue you. This was a really dangerous place to be investing capital. And that fall off there for July, that's a, obviously, a partial month. So that might only be a week that would probably explain that drop off. If you look at June, $5 billion raised,

that's a big number. I wonder if that includes circles. IP I would think it does. So, and then the other $5 billion, I wonder if that's another crypto public company as well, or if this is all just venture dollars into those companies. This is all venture dollars, as it was described by a crypto rank. But, you know, everyone counts slightly different than Jason. So I don't want to be overly proscriptive. But my read was this is just venture. So it would not include the circle round. Yeah. And these rounds are also very strange. Like these could include ICOs and people selling coins and then getting back other coins. So I don't know if this is all venture capitalists putting money. in, but let's put that aside for a second. I think you could feel pretty good going to an attorney and saying,

if I'm going to do this stable coin thing, if I'm going to do this, will I be sued? And they would say, a lot less chance than under Biden. So if the president can launch a meme coin, I think you can too. And I think that's kind of where these, well, I mean, let's call it what it is. You know, they've said meme coins, NFTs, these are collectibles. If you want to buy collectibles and they trade with a table, or symbol on, you know, trading accounts on Robin Hood where I bought my Trump coin and I've lost

70% and I did it as a joke, obviously. But, you know, you, this is now the rules of the road. And I think it was previously, if you do something that's innovative or new, you're going to get caught. And it's just a matter of how big the settlement is. And if you can survive. we can see XRP as an example of that. They were able to survive. They had enough money, enough cash that they were able to fight that lawsuit until Trump came into office and XRP was, I think, is largely off the hook now. I don't know the exact state of all their cases, but I believe they won one case.

They weren't able to prove that they were doing solicitation of non-accredit investors because those things were trading on marketplaces not being sold directly. by XRP. So it wasn't that they weren't selling to consumers. It couldn't be proven that they were selling to consumers. Putting all that aside. I think now people are, you're in the clear,

is probably how most people feel to experiment. Does it mean commit crimes? Does it mean upscone with the money? Like we saw with ICO. So I think we'll see a lot of innovation. I've always liked fintech. We were involved in Robin Hood and Wealthfront. I think fintech's a great space, you know,

and adjacent to fintech is crypto. Sometimes they overlap. Sometimes they don't. But now with crypto regulations being out there, I think you're allowed to dream again, hey, what could the world look like if you could tokenize shares of a private company like Robinhood did with Open AI shares?

Like, okay, it seems like there's enough legal consensus that you can experiment again. So here we are, folks. Great time to give it a shot and invest. Everybody's got to take as investors and as founders, you got to define your own level of risk. And when you talk to attorneys, they will almost universally tell you,

like, we can't give you a legal opinion on this. It's too new in these kind of areas. And so then they rely on what's happened to other companies and other cases out there. Well, if all the other cases are being folded and settled, okay, yeah, they can tell you we don't have exact law or, hey, the stable coin laws

are in process, this is where we think they wind up, here's what Circle did, they're in the clear, and then if you wanted to innovate beyond what Circle's doing and maybe get closer to tether and offshore marketplaces for crypto, etc., and exchanges, yeah, then you're going to be taking more risk.

And jurisdiction is the other piece that really matters here. If you're doing stuff in Panama and Malaysia and Singapore and Dubai, like Zerg, different places have different rules. The United States is catching up up to those. Yeah, but catching up by having fewer, I think is maybe the right way to say that kind of a race to a more relaxed market for entrepreneurs. And that's why when I see companies like Agora raising for its, you know, stable coin product, well, it kind of just tracks with how you and I view the world and what we see coming. People want stable coins, they want to be able to

trade. Okay, well, off we go. Now, before we get to our guest Jason, a couple of small things. one, GROQ, GROQ, not GROK, the AI model. This is the inference provider, of course. I believe Sonny from our old AI shows. Yeah, he's now, he's part of the GROC family, right? Yeah, he sold his company to GROC and he is responsible for all the developer relations, which I think it's like Developers.crock.com, GROQ, as opposed to GROK, which is Elon's GROC. These folks make inference ships, and they've done a ton of different deals.

And they're raising more money, it seems, yeah. Yeah. So according to reporting, they're going to raise between $300,500 million at a $6 billion valuation. Well, for context, their last round, Jason, 640 at a 2.8, so a more than doubling of their value. Why? Well, they signed a big deal with Saudi Arabia.

And I believe it was the humane project. Their AI had a G-42 competitor. They also announced an expansion into Europe. And perhaps most importantly, the information reports that they grew their revenue from 90 million last year to an extent. expected 500 million this year. So 6 billion, 500, 12x makes a lot of sense to me. I don't know their margins. I don't know how repeatable their revenue is. We can't say it's an ARR multiple, but 12x does not see nuts for that level of growth. And I just love to see companies that are

not just Nvidia, but our building chips for the future do well because back to our point about AI models, steel, sharp and steel. And we want to see a lot of companies pushing ahead here with new ideas and so forth. And GROC makes LPUs, not GPUs. Those are language processing units as opposed to graphical processing units. So I'm really excited about this. I'm keeping tabs on GROC's. I think it's a really cool company. And I just hope that they do really well and make sure that Jensen stays up at night. Yeah. I am a shareholder. And they, of course, I was a shareholder in Sunny's company which got bought. So I have some modest exposure here. So apparently this is good for me. No, I hope GROC crashes into a wall and I hope not. I'm just kidding. No, it's so far so good.

you know, the inference chips are, is there a different architecture than, say, when you're building your language model and you have to compile it essentially. So there is going to be an infinite, it seems, or in the medium term, an infinite appetite for this as consumers stop doing jobs and businesses stop doing jobs inside of apps or SaaS software

and they start with a language model. And when you start with a language model and you need a context window, You want to throw a video in there and say, tell me about this video or a book or a book and a video or you want to index your entire hard drive and make that the context window. Like, it's going to be pretty crazy.

I don't know when we get to the point at which this becomes like storage. And people are like, I don't know how much storage I have or need because every three or four years, like they keep giving me the ability to record in 4K, 8K, whatever. you know, and doesn't make the videos any better. And the storage keeps going down as our phones and the quality of the video goes up. So even as we make bigger and bigger files, it doesn't seem like the storage business is a great one to be in.

At some point, 10 years from now, 20 years from now, we'll be looking at these, this kind of infrastructure in that way, I'm sure. But that's not going to be for another 10 or 20 years. It took, gosh, I'm trying to think about how long we sweated storage. on computers for consumers or businesses. And I would say for consumers, it ended sometime 20 years ago, where you really didn't have to think about it, right?

Like, you weren't buying your computer and saying, how much storage does it have? Because you had cloud become available, and cloud was easily expandable. So that was 15, 20 years ago, 2005 time period. RAM and processing power of chips, gosh, I think that stopped

maybe 10 or 15 years ago, where people don't even look at how much RAM they have on their computer unless you're doing video editing or video games. No. Or you're a tab hoarder in Chrome. Because Chrome, Chrome's, I mean, you're right, apart from Chrome. Yeah, but you don't need to have that many tabs open. That's like a choice.

That's like a crazy thing. That's like, I don't know. I'm going to make a two-hour, 8K video on my, or 4K, I don't know if my phone can do 8K. He does 4K for sure. I'm going to make like a 60 frame per second 4K video of like a flower growing. Like you don't need to make that. You can make a short 30 second video and it's the same thing.

It doesn't change. You're choosing to open infinite tabs and then complaining when you run out of memory. It's stupid. It's like it's unnecessary. Just close your tab. In fact, people like you are now causing these browser companies to add features where they turn off and they shadow turn off your tabs. Like you know that's what's happening now. You're welcome.

turn them off. Well, you can also close your windows. Why do you keep all these tabs open? I don't get it. Well, I mean, just for this show, I have about 15 articles pulled up for various contextual points that I might need to reference. 15's fine, but you're...

Then I have another entire... Sorry. Go ahead. You're... We all know 15's okay, but I think you've got 50 tabs open right now. Oh, yeah. Then I have a separate Chrome instance

that has all my screen shares prep for the show. And then in my third Chrome window, and then on my other screen, I mean, we pull together a lot of stuff for this show. And so I just, I use a lot of tabs. And wait, how do we end up here?

Oh, RAM. Right, let's get back on topic. So RAM, back to RAM. Like, do you possibly need more than 16 gigabytes of RAM on a laptop if you're not doing video editing? I think the answer is no. And I have now stopped even,

I literally get the lowest amount of storage because everything's in the cloud. So I don't think about storage anymore. I don't think about, the CPU or the CPU anymore because they're all off the charts in Mac and on PC. There's no game or something I'm playing that needs that. And then even video editors, they used to come back to me every year and say, oh, you know, can I get an upgraded computer? Now they have computers that are two or three years old. And they're like, yeah, it works fine.

Works fine. I'm like, should we get the new one? It seems like it's faster. It's like, yeah, it's 20% faster. We don't need it. So that's going to be interesting to see if that happens with AI clusters. And then you think about like Colossus that Elon built, that really, I think is a big part of why Grock hit these things, is that they have this giant cluster. How long are those GPUs good for? You put up 100,000 of them.

You can use them for three, four or five years. What's the lifespan of those? Microsoft recently changed the accounting of its depreciation, Jason, pushing it out by one more year. I think it was four to five, from four to five. So call it a half decade. Okay, five years.

So you spend all of that money, billions of dollars, 100,000, you know, per GPU or server, I guess. You know, like, and then it's gone in five years. And then what do you do? What are we going to do with all these colossus data centers? Do they stay online? Do they have some afterlife purpose? Or do you, they're not worth running anymore

because the new ones are so much more energy efficient that we just turn them off? or can they be used for something else? I think it's like actually, if we're going to build out this much infrastructure, people should be asking these questions. If somebody, let's put this on the docket for Monday, I would like to get the answer to this one is what happens?

If these things do have a five-year life, what is the post-five-year plan? What is the plan for year six to ten? Do they get turned off because the energy consumption makes them not worth using? Do they get resold? Is there somebody who needs to use them? In a world where AI has solved all those problems,

then you don't need these chips anymore. The problem's been solved. It's been baked into an LLM. I'm just really wonder what happens to all these. I wonder if Nvidia has like a recycling problem. Because with Facebook, how long does Facebook keep their servers on? You know, in that open cloud, they had that open computing standard they created. I wonder how long they depreciate theirs for before the energy consumption of them

in ratio to what they're capable of doing becomes untenable. and you're just better off hurting them off. And then if you turn them off, is it worth taking them out of the rack for the real estate? Or do you just leave them in the rack and then just build more racks behind them in a giant data center and just turn them off?

Because the real estate cost is less than the cost of dragging them out of there, depending on where you are? I have a lot of questions about all this. So we'll talk about it more, but just a couple of quick data points. According to a number of sources here, when you do run these GPUs in a day,

Center at like max capacity for a while, they do degrade. So they do actually have a life expectancy to them. Now, the major companies have pushed out their depreciation calendars up to five. They break? Or they are like a battery on a Tesla or an EVA that they degrade and they capable of 60 or 80 percent? I wonder what they mean by degrade, like actually break or they perform at a lower ability? Well, according to training data from META's Lama 3, there's actually a 9% annual failure rate

of GPUs used for training information. So again, we'll do more research here, but that's one thing. Wow. Okay. Yeah, that's a lot. So that means every 10 year, or every 11 years,

they have all failed? Wow. That's interesting. That's brutal. But I think the real question is, how fast does NVIDIA push the envelope out? Because every year, if they have a faster chip out,

that's much better, I think that makes, yes, last year's chip depreciate faster. So if Vindia runs into a problem that it can't expand capacity, or performance faster, then I think these last longer. But if Nvidia is better, they get shorter lifespan. So we'll talk more about that on Monday.

I'll do some research and touch some people. Yeah, and the other piece of research is, I wonder what it costs to ship them and to rack them and to unwack them. Now, we have a guest, Jason, so I'm going to go ahead and move our click to cancel work until Monday. There's a lot of interesting stuff there, but it's just going to take too long to get us through. So I think it's time for a quick commercial break while we get our friend into his chair. So Claude is not just writing show notes for us anymore.

It's gotten a lot better. It's actually analyzing guest applications. It's helping us draw up interview questions based on founders' specific background, what they're doing. And it actually helps us generate research in real time as we do the show. And that's why Claude is perfect for startup founders. It is intelligent. It is quick. And it will help you matter what you are doing.

Now, if you do create a lot of code, of course, Claude code is an excellent pair programmer for you. It can help you out even, I believe, in your terminal as well. So no matter what you're building, well, Cloud probably has a solution to it. And if you need to bring your own data into the context, well, don't forget, there is Model Context Protocol or MCP

from Anthropic, the makers of Claude. And what that will do is ensure that your private data, your proprietary data, the stuff that you need to become great in the world of AI is available to your favorite AI model. And that's why we have partnered with Anthropics Twist listeners, 50% off Cloud Pro for three months. if you go to clod.a.ai slash twist.

We know if you're bootstrapping out there, every dollar counts, so let us save you a couple of them. That's clod.a.i slash twist. All right. Next up on the program, we have a founder. I'm a second time investor in Ben's company.

His first company, neighborly, such a great idea during COVID, and it wound up shutting down. We'll get into that. But you have a new company, and it's Ben Sidel. I'm pronouncing your name, correct? You are.

Settle. And the new company, you came to me and you pitched me this idea. I said, that sounds crazy. There's many more questions than answers for this one. I'm in. So let's talk a little bit about what you're building with Autolain. Yeah, so with Autoling, we're building the connective tissue between autonomous vehicles and businesses. So it's essentially an operating system that allows high volume retailers, shopping centers.

So think of like a Whole Foods or a Costco or. or shopping center, like the domain here in Austin to be able to connect, coordinate, and control autonomous vehicles on their property. Got it.

So I have something like the domain, which is an outdoor mall here in North Austin. It's pretty impressive when you see it. It's like you can live there and there's an Apple store,

tons of restaurants. They want to get deliveries there. It's going to be automated, obviously. There'll be self-driving cars. We've got self-driving cars all over Austin for whatever.

reason this has become ground zero, it seems. But there needs to be some connective tissue. When you're in L.A., there are those street, what do they call those little robots on the street? Oh, the sidewalk bots? The sidewalk bots. We're seeing Zipline in Dallas doing drone delivery. But the big opportunity in your mind is a self-driving Tesla, a Waymo, a Volkswagen ID buzz. A zooks is going to show up at the domain, and somehow at the domain, a retailer is going to take a burrito or a pair of sneakers and put them into a self-driving car that's going to come to my ranch, out in Hill Country, and give me what I want. Yeah? So which piece of the puzzle are you going to take care of? Or is that all the point of this startup is to figure out, what needs to get done? Yeah, so we can't put the idea, essentially,

if you look at the components that you would need in order to make that type of a transaction possible, of course, you need a self-driving car, I need that technology to be flawless, and I think we're pretty close to that. So I kind of envisioned, well, fast forward a few more years, and let's say most OEMs are either licensing or shipping some type of personally owned autonomous vehicle.

Toyota, GM, Ford, you'll have a few more. bunch of these companies shipping these types of cars. And if most of us have access to this technology, don't these cars essentially become physical AI agents? These become your kind of like Butler, right? Yeah. So in that model, I have the car. I have a model Y, hardware four, pretty incredible, not perfect yet, but getting there, pretty close, right? I think I would say two out of three drives I do is flawless. One out of three, I can have a,

that was suboptimal moment or, oh my God moment, to be totally candid. Like when it gets near construction, that seems to be for these cars really challenging. But I might send my car to the domain and say,

hey, my car's there. What now? Yep. And that's, and that's, so that's what we are building. So we,

most of the, I would say, autonomous driving technology has been applied to ride hailing thus far. So, of course, Waymo has been operating in public for years with almost no significant incidents. So I don't think there's been any, nobody's gotten hurt. Nothing. And so, you know, they've, they've now have more market share than Lyft does in San Francisco and, and still growing. So I think that San Francisco, if you were to look at the market share for ridehaling, you could pretty much assume that what you're seeing in San Francisco is going to continue to expand and duplicate across the country over time. Now, it won't

be an overnight situation, but of course, AV ride hailing is here. Yeah. And if that's any indication of where the general market is going for TNCs and aggregators, you should expect that delivery will also go that direction over time. And so what we've been building at Auto Lane is essentially the connective tissue to say, if that is to take place, the vehicles are here, the AI driving systems are here,

the regulations are coming, consumer adoption is happening, but if you've taken a Waymo or you've used FSD to go to a business, you'll quickly notice that one of the hardest parts and kind of least developed parts of the technology is the pickup and drop-off experience.

Yes. And because of that, and that is what we call the last 50 feet, because that is the part that is really unrefined at this moment, that's where we are focused on at Auto Lane is to build the software and hardware to basically enable physical and digital infrastructure to make that final 50 feet polished and flawless.

And then if that can take place, then all of a sudden you have the ability for these cars to conduct commerce. And that's where we think that the next wave of commerce is going. I really believe that we're going to see something akin to the transition we saw with e-commerce. We're going to see something that I would coin autonomous commerce. It makes sense to me because there are places I've wanted to go here in Austin. There's like a sand-dow shop. You know, Japanese sandwiches. And I have wanted to go to this place in North Austin,

but I live in South Austin and the whole country, and I just haven't had a chance to go up there. And there's traffic, et cetera. Now, if I could, I would have ordered from this place every week, some sushi or a nice Katsu Kari or a, you know, wagu sandwich on that great Japanese bread. Yum, yum, yum. It would have been very easy for me to just say, I order it. My car will be there, just put it in the front trunk with a nice pack,

and yeah, my car comes back here to Capital Factory where I have our studio, and I jump back in my car and go home. That's right. It could have done that errand for me already. Yep. And two or three other ones. Yep. But there is so much logistics in what I just described. That's right.

Which is why I was really excited to invest in the company. Yeah. And thank you for that. Oh, well, of course. I mean, the greatest success we've had is when we've known a founder for two or three startups. and your first startup was amazing. I still think it's one of the greatest ideas I've ever heard, and I still think it will work.

But I'm glad you went on to this one, because this one feels like it's even bigger. And that's typically what happens with founders. They have some amount of success in cigar tissue in the first one, and the second one, then they're just like, oh, yeah, I'm not going to make those mistakes. I'm not going to overspend on this. I'm not going to underinvest in that,

and they just, boom, boom, boom. So we just love second, third time founders here, especially if the first company or second company was a complete disaster blew up or was like a modest success because those things kind of set you up and you still have the chip on your shoulder. We'll get to that in a second.

Also, this is going to be super fragmented is your best guess. Oh, yeah. Super fragmented. So explain the fragmentation and then what you perceive a retailer, what set of problems is the retailer going to have

due to that fragmentation? Great question. So, yeah, one of the easiest ways to think through how this industry could play out from a solution standpoint. So let's take Walmart, for example, okay? So if you agree with me that personally owned autonomous vehicles or autonomous vehicles in general as a technology are going to scale over the next five to 10 years, if we agree on that? Everybody agrees. Okay. Well, then I guess you would also need to agree that those cars are going to go to Walmarts. Yes. They already are. Yeah. Waymows are dropping people off,

picking people up from those stores, or isn't commerce taking place with Walmart directly yet via those vehicles, but of course there will be. And so one of the major aspects of this conversation that I think is important to consider is whether or not, because the industry is a bit divided on this topic, but whether or not you believe in personally owned autonomous vehicles

being a significant market share within the autonomous vehicle industry. Some people believe, oh, the technology is going to scale and the costs are going to come down to a point where OEMs will operate their own robot taxi networks and people just won't own cars. Some people will choose not to own a car. I think there's a lot of people who believe that side of the spectrum of the industry where essentially you won't need a car at all. And the services, both ride hailing and delivery, will be so cheap

that really won't be a need to have a car. So maybe that becomes the case for some people. For people, I could see it, for people who live in an urban environment and don't have access to a garage. People are so that about Uber, you know, sincerely days. And it remains to be seen whether that played out the way. Well, Uber, I would say, is like 5% of the population. Yeah, exactly.

I don't think it played out the way. Totally. And so I think, we believe a similar situation is going to arise here with autonomous. And actually, it's not, in my opinion, the case that we just won't own cars. It's that actually the cars that we do own are going to be able to do a multitude of things for us and they're going to be much more valuable and we'll need less of them. But I do think most people will own a personally owned autonomous vehicle in their driveway and it can do errands for them, it can pick up their kids from school,

it can drop them off at the airport and drive home, can do all types of things. And so we might, instead of having two or three cars as an average American household, we might just have one or down to one and a half. Because when I get dropped off here at work, I could send my car to go pick up the kids from camp. The car, then they could also make their way to, I don't know, back to the ranch to run some errands for the housekeeper and get some chicken feed or something. So we really think that personally on autonomous vehicles is going to be the future transportation. Yeah. And if that's the case, and you go back to kind of your question about Walmart and their fragmentation, so if most people have access to this technology, whether you're on the car or not, or whether it's your neighbors or your friends, and you order something from Walmart, groceries, retail, what have you, well, that car is going to be able to go to Walmart. Now, if that's a Toyota or a Tesla, or that's going through DoorDash, or that's going through Uber Eats, or that's any, or that's a drone, I mean, you're talking about, Not just cars, but all types of robotics here that are going to do consumer transactions.

And so if those vehicles are going to Walmart to do commerce on behalf of a consumer, imagine being Walmart in this case. Yes. It is a complete nightmare that is about to unfold on their hands because... In their parking lot, there will be 10 times as many cars showing up. That plus there will be 15 to 20 OEMs. And each one of these OEMs have a... different hardware, software, and data stack.

Right. And that's not one type of vehicle. They have to already design their parking lots for a number of vehicles. You could have, you know, trailers showing up in the back to unload. You could have, you know, SUVs. You know, so right now a parking lot has compact car spots, spots for the trucks, spots for SUVs and larger ones, and maybe a pickup, a drop-off zone. They've got maybe arguably...

And curbside pickup. And curbside pickup. So they've got maybe five areas right now. now you need another area where my self-driving model Y or your self-driving Toyota because Waymo just said they're going to license Waymo's

self-driving to Toyotas. You saw that announcement, I'm sure you're on top of that. Hyundai, Hyundai as well, yeah. On Hyundai, they're doing it. They've got a partnership with Hyundai. Wow. So all those cars are going to have self-driving. They're going to show up in a parking lot.

What parking space do they go to? Yep. Do they go to the pickup zone? Do they go to any spot in the pickup zone? How do you know the car is there? How do you know where to put the stuff? Is it go in the backseat, the trunk, the trunk? How do you unlock the trunk?

How do you unlock the car? How do you know which VIN it is? How do you know which customer is associated with which car? How do you know which stall that car is in? How do you know which order? If you have three orders at Walmart and my car goes, well, which order is it? There's basically what I call the three Cs of AV orchestration that Walmart or any major retailer is going to need to solve.

One being communication. So this is Jason's model. why, it's coming to pick up order number one, two, three. It'll be here in 16 minutes. It's license plate, blah, blah, blah. It'll be installed two. Okay, all that information needs to be communicated. The second part of is coordination, making sure that that car is going to go where Walmart wants it to go because the curbside pickup operation across 4,500 stores, operating 16 hours a day is a very expensive labor line item on their, on the P&O. Yeah.

So when they can shave off 15, 30, 45 seconds from a pickup and drop off curbside arrangement, then they are operating more profitably and then they can knock the prices down on their retail good, which is the entire Walmart business model, right? Operating more efficiently, reduce the prices for consumers and sell more goods. And so they are obsessed with finding ways in which to do these types of things more efficiently. And so is McDonald's, so is Chick-fil-A down the line, H-E-B. And so really what you end up with on those three seasons, you've got the communication, you've got the coordination,

the car needs to go right here so we can be faster. And then the last part of it is control. Now you don't need to tell the car how to drive to Walmart, but you do, when it gets to your private property, you do need to say, okay, well, I need to be able to unlock this car, unlock the front or unlock the trunk. And if you have 20 OEMs shipping different types of autonomous vehicles, you're Walmart, you're going to have to have an integration

with every single OEM. And those OEM integrations, they're not stable for five years. Tesla's APIs change every two weeks. I know that because we have integrations with Tesla. So you're going to have 20 OEMs that are changing their code base every two weeks. That's an entire in-house team

that you're going to have to connect with cars. Potentially could build this software, maybe, but they're one of the few who could actually create this. And then, sure, Toyota, Hyundai would have no choice. but to interface with Walmart. That's right. But after Walmart, that drops off pretty quick.

It does. Yeah. Like, even H.E.B. is a local retailer. That's right. They're not across the entire United States. That's right.

Starbucks, yeah, they might be able to figure this interface out with their drive-thru, maybe, but still, you know, with all these different players showing up, I could see a world in which 10 years from now

there would be more autonomous vehicles in a drive-thru, in a Walmart parking lot than actual huge. human customers. All right, everybody, and this has been, and go check out Auto Lane. Is it Autollane.com? Goadolane.com. Go AutoLane.com. And we'll see you all next time on this week in startups. Bye-bye.