Anthropic + SpaceX, Codex 100x'd Claude Code & Half Cash Half Stock | GMP EP12

Half Sock, Half Cash. Did you see my meme by the way? So, the beta meme, Half Cash, Half Sock. Yeah. Then I change it to Half Cash, Half Pork. Get the MRR and then we get the mic. Cool. What do we expect? Honestly, I looked at all of the latest thumbnails and they're looking super hot. You must have the same experience where your feed is just AI slop, like constant AI slop. But you guys, it's actual your takes on what's going on. It's not AI rubbish. It's just, it's nice to have that on my feed. We had our first video that hit 100k views because of the contrarian take of Codex took over Claude. Why do I always have to make this Barcelona versus Real Madrid? Coding in App is the new starting a podcast. We are three dudes. We just started the podcast. It's incredibly reckless in this world to not learn AI. Everything is code. I recently realized everything is code. Good morning, good afternoon or good night from wherever you're watching. This is episode 12 of the God Mode Podcast and I'm repeating myself again, but we have the fullest docket that we've ever had. AI land is not sleeping and there's some crazy stuff happening. In the beginning of the week, I was messaging my friends here that Antropic peaked. I thought Anthropic peaked, Codex is taking over. But a few days later, actually, we have Antropic shaking hands with Elon and yeah, feels like the run just is going to continue. We're going to talk a lot about this. We're going to also talk a lot about news that came out from OpenAI, from Cursor, from Elon, from SpaceX. Those are the main topics of the day again. But hey, we also have a friend of the pod here and he's going to talk with us about agents, especially Hermes agents. Alan has been building agents since before this podcast existed, which is quite a long time in AI land. I actually saw today, I think he posted that he's using his agents to learn a new language. So let's get started. Alan, maybe a quick one on you, a quick introduction and what are you using your agent for? How are you learning through your Hermes agent? Yeah, thanks for having me on. I have been using agents, like I think ever since OpenClaw went viral, like my folder, which I use my codex and my code code to build it. That's called MaltBot. So you can imagine like that's like ancient history. It's like back in the day when it first went pretty viral. For me, I think about my background, I'm a software engineer. I would say I'm more of a tinkerer. I would say I'm just like experimenting and building stuff. So some of the stuff I have you built with the agent, some of it is useful for other people. And a lot of it's stuff that I just love to use in my day to day life. So yeah, you mentioned today I built a language learning agent. So I like getting AI feeds, interesting topics from X and I get summaries of that. And then I just have my agent call in that information and I use like a Google model to get that transcript into audio. And then I have like a listening class in Mandarin. I used to live in China a few years ago. Yeah. Okay. So you're translating the general news on X into Chinese. Why don't you pull the data from a Chinese news source? Because I think it's even more interesting to learn what's happening in China. No? I would love to know what's happening in China. I think that doesn't really reach our feeds. Yeah. I mean, there are Chinese users that use X as well. They use their VPNs to get through. But yeah, there's other platforms that they use like Little Red Book and there's other users. There's actually an influencer, a Chinese influencer who's here who works in the social media space. And she gave me a free KimiK 2.6 early access key as well. So there is a lot happening, especially in the open source side in China. Right? It's probably the most important, like interesting thing that's happening in China. And I have some, I lived in China for like three or four years. I haven't have left and don't have as much connections there, but this is what I want to do. I want to get the connections back. I want to see what's happening in the open source space because the best open source models are Chinese, right? They're not racing that ultimate race for AGI. They're trying to get cost optimized models that everyone can use. And it seems like in terms of business perspectives, that's what everyone kind of really needs for customer optimization. They're much more interested in that. And so I just, what I wanted to do is my feed is already chocked full of AI. I told my agent, I want you to pull in topics that are Chinese, like Chinese language users, small accounts, interacting with big accounts and big accounts as well. And then pull that in and give me the summary of it, some vocab, and then like give me a transcript with, you know, using a scale to just give me a structured way to have a listening class. And that's all just, yeah, within like 10 seconds. You mentioned that you're using, are you a Claude fan or a Codex fan? You're using Claude code or Codex? What's your go-to? So I have both subscriptions and I've flipped between. I was always the biggest Claude fan, I have to say. But recently with the, I mean, I saw your video that you made with the rate limits, with the controversy with OpenClaw. And now that you can use OAuth through Codex, just with, you know, the rate limiting and the user experience, I decided to use Codex last week, a few weeks ago. And my user, like the speed, the sharpness, like I feel like the latency of the requests, everything just feels much sharper with Codex. So recently I've switched and I use it. I now use Codex every day. And I'm a big Claude fan, but I think just with what's happened in the last few weeks, I'm moving over. So I'm one of those that's downloading Codex. Yeah, it's funny because, and that's what I mentioned in the intro as well. In the beginning of the week, it really felt for me the same experience. I spent the first two days of my week in Codex and I barely touched my Claude code. And I was even kind of sad that a few days ago, my Claude Mac subscription re renewed because I would not have renewed if I found Codex earlier. And like the stats here show exactly the same numbers where Codex has overtaken Claude code in downloads and it's not even close where Codex had 46 million downloads, where Claude code only had 491,000 downloads. That's, that's like a, how much? A hundred X difference, which is pretty insane. And based of these stats, I sent, sent you guys in the WhatsApp, I think Entropic peaked. There's no way back here. Guys, have you, have the rest of you been playing with Codex yet? Ben? Uh, yeah, no, I haven't been playing with it yet, but I, um, I've been going to this like weekly meetup in Valencia with other hackers. And, um, it seems like every week there's like one person who's gotten Codex pilled and, um, have made the move over. Um, one interesting thing that I, that I did kind of hear this week about the difference between Codex and Anthropic, which is related to this, but it's kind of, uh, an interesting, uh, design decision that I think, uh, open AI made. Um, which is surprising to me why, you know, Codex is still taking off basically. Okay. So to GPT-4O people loved GPT-4O because there was a personality to that model. Um, but then apparently this week, someone kind of reported that like open AI made a very intentional decision with GPT-5 and like the five models to strip away that personality aspect of the model. Um, and so that is probably why a lot of people prefer, I think, uh, Claude when it comes to, um, personal assistants, right? Claude code or Claude for, for personal assistants, because you want a personality with your personal assistant. But when it comes to coding, you might not actually want that kind of personality, that character and GPT-5 doesn't have that character. They made that decision on purpose, right? Open AI did. Um, and so perhaps that's why people like Codex is because like, okay, the model just does the task that you're asking it for. It's not pushing back. It's not kind of asking questions. It's not, um, you know, doing these things, which is great when you're building something like Hermes or OpenClaw, but when you're coding a product, you just want it to do the thing. Like what, when do you decide to use Codex versus Claude code? Um, like what, what type of task will, will you say, okay, this is better for Codex? I think, uh, we have a team who, who use both, uh, the company I work for. Um, and I think that the guys who, who do a lot of front end work, they like Claude more for that. Um, the thing is in my view, if you use the right prompts, if you use the plugins for good UI, um, you can, you can definitely get as much good out of Codex as Claude. I think it goes back to what you were saying about you, like, you just want to get the job done. And in my experience, um, when I use, when I use GPT, when GPT 5.5 came out and I started using that, just the, the ability and the sharpness of it just felt a cut above Opus 4.5 and 4.6, or now 4.7. I think the token economy use of GPT 5.5 combined with the fact that I think, I think, I feel like Claude code Opus is more of like a startup. Let's get it done quick. It likes to cut corners. It might do things, but might not do it in the most, in the best way. Whereas it feels like chat GPT will like the, sorry, the GPT 5.5 will, will make sure you do it right and high quality um, from the start. And I just like that about, about Codex. So when you combine all that with the issues, with the bug that Claude code had and the rate limiting, I just had to switch. Yeah, that makes total sense. Especially the token economy. It's like, it's not really how many tokens in and tokens out. It's also how many tokens you actually need to get a task done. And I feel like, um, in that sense, GPT 5.5 is more efficient to actually get something done in a, in the right way, especially compared to Opus 4.7. Is that in line with like, what you mean, Alan? Yeah. I mean, I know the token burn is much higher for, for Opus. Um, but I think, I think it's the rate limiting. I feel like I can use my, like a $20 subscription with Codex. I feel like I get like as much as a $100 with Claude code. And I think that's even after they said that they've fixed the caching issue that they mentioned at Claude. So I don't know exactly what's going on, if there's still an issue with that harness or the, it's just the, the, the model itself is much more efficient, but that's, that's my experience of using both. Well, um, talking about rate limiting, I think these, this might've been the end of rate limits at Claude from now, for now on. Um, because I think on Tuesday, two days ago, um, Claude announced that they agreed to partner with SpaceX, which will substantially increase their compute capacity. This along with our, with their other recent compute deals, um, and they've making a lot means that the, they increase the use usage means that they are able now to increase the usage limits for Cald. Oh, what the fuck is going on here? Um, this basically means that we won't be rate limited anymore. Um, and that we, that our users usage limits basically to X now. Um, and this is the second tweet that they wrote. They're doubling the code Claude codes, five hour rate limits for pro and max and team plans and removing, removing the peak hour limits. Um, and I've been really noticing the difference when I'm using Claude, but this is a crazy partnership. Um, in my opinion, and I'm not sure if we have been talking about this before or speculating about this, but Shamat was right. Shamat mentioned, um, that Elon should just partner or that Amtropic should partner with Elon. And yeah, they did. And this is, this is very impressive. The, the one thing that, um, um, Antropic was lacking to actually continue their growth was compute and was power and was energy. And now with Elon together, they, um, yeah, I think they can, I can continue their, their journey up and their generational run where, um, yeah, where Antropic was dead on Monday or peaked on Monday on Wednesday, they're completely back. Ben, what, what do you think here for you? This is awesome news, right? You downgraded from 200 to hundreds. And now with your hundreds, you probably can do double as much. Yeah. Well, no, I, I, I, then last week, like two days after I went back to the 200 cause I, cause, uh, cause I hit that rate limit. Um, so, so I'm back on the 200 and, um, yeah, no, I, I, I, I looked at some of Elon's tweets and, um, um, it's interesting. He, he, he, he very specifically, uh, wanted to caveat that, uh, Anthropic was had access to Colossus one, right? So Colossus one was built a couple of years ago, I think two years ago, and it's about a hundred thousand to 200,000 of the, like the hopper, the, the, the previous generation. Now XAI and SpaceX have moved on to Colossus two, right? Which is 500,000 of, of the newer, um, you know, black whale chips and all of that. Basically Elon's giving them like the, the last generation kind of the older, uh, data center. And he's moving on to the new one, which is like three to four times as, as powerful. Um, it is interesting that Elon's like striking up all these deals, whether it's with cursor, whether it's with Anthropic, like I'm wondering if SpaceX and XAI, like they've just, they massively overestimated how much compute. And they were just, they, they were the only company in the world that basically had all like thousands of these data center racks just sitting empty. Um, and they maybe didn't even know what to do with them. I don't know. That's crazy to think. Cause like everyone is, yeah. What do you think? If you, if you listen to some of the recent interviews with Elon, he is saying energy is the currency of the future. So I think he has seen this all true and he just a few steps ahead of everyone else. And he invested in the future and he knew that compute is not going to stay unused. Um, and he might not be winning the AI race, but this, um, with, with Grok and XAI, um, for whatever reasons, right. You cannot win on every front, but now he literally has a partnership with the best model creators in the world and they will continue to work together because the next tweet here as well is, uh, SpaceX and Antropic have also expressed interest in partnering to develop multiple gigawatts in orbital AI compute. This is, I mean, this is the next level, right? This is, um, the energy solution for the future. And if Elon can do this together with, Elon can do this together with Antropic, then does he actually still need XAI or does XAI actually just become a provider, uh, for Antropic? Uh, what do you think? So, yeah, yeah, yeah. I, I, I, I agree. And I, I think this is where Elon's getting a little carried away because on one end, Elon hates Sam Altman and he has like a personal vendetta against him. But over the last month or two, it's become increasingly clear that Antropic is the more closed, uh, AI provider out of the two, right? They're, they're limiting what you can, what you can do with the models. They are, you know, cutting off the department of war and making very like harsh stands on like what you can do. So, so Elon, I don't know why he's, um, you know, I know why he's partnering with Antropic, but he's doing it not from like a, uh, and, uh, uh, uh, uh, a theoretical standpoint of, of, of, of I align with how Anthropic is running their business. Like even three months ago, he was mocking them. He was calling them misanthropic and all of this. He just literally is doing this because he has the compute and this deal is probably paying for all of the, um, training of the Grok 4.5, Grok 5, that's going to come out and all. So I get why he's doing it from a, from a financial standpoint, but you know, Elon is someone who I at least feel does have a fairly solid, uh, moral spine when it comes to like, you know, being truth seeking and all of that. And for him to hand all of the compute over to Anthropic, which is a company that frankly is pretty close sourced and really cares about like, like, you know, they're almost deciding within the company, what, what is acceptable and what's not in this AI world. It doesn't really align with like Elon's kind of morals when it comes to AI, you know? Yeah, but I've also seen some tweets where, um, where they say indeed that this is a pure win for XAI and Elon, like, um, Anthropic is going to pay them to use that compute, right? It's not that they're giving it for free. So, and then like, Elon is thinking about, I want to create a million dollar, uh, no, a million people colony on Mars. Like you need funds for that too. This, and Anthropic is the fastest growing company in history, right? So like if they continue growing and they continue paying Elon, then Elon has enough money to fund everything. And 10 trillion or bust is what he tweeted recently. Well, um, I think this deal can put him well on the way to actually reaching that 10 trillion dollar mark even sooner than, than later. He said that, he said that 10 trillion or bust for his own net worth. Is that what he said? Yes. Yes. I'll find the tweet, but Peter Diamantis basically tweeted, um, that Elon, that Elon currently basically is worth about 2% of, um, the GDP of America. And the last time someone actually did that was Rockefeller. And then, um, Elon commented on that tweet at 10 trillion or bust. God, that's crazy. One thing I keep in mind when I'm seeing these new plays coming along is that these are all three individuals with huge egos that want to be the best and win. And I feel like Dario has been put in a position where his hand has been forced because Anthropic started falling behind on their compute. And the recent acquisition of cursor makes SpaceX not only a compute provider to Anthropic, but also Anthrop one of Anthropics' largest customers. So whether Dario or Elon were on good terms or not, I think in this case became almost irrelevant because their companies are so tightly linked and by teaming up, I guess, yes, they could beat, um, their mutual nemesis. You know, in this, if you looked at it from, uh, an art of war perspective, the enemy of my enemy is my friend. Uh, so I think that's where we're at with this partnership. Exactly. Alan? Yeah. I'd also say I wasn't actually too surprised by the, by the partnership in some ways because compute is so constrained, right? So compute is what, what all these providers really need. What I thought was most interesting was, was what you pointed out about the orbital satellites, where we host the future infrastructure, where all the inference will happen. Because if you look at, if you watch a lot of interviews from Elon Musk, why, why is he saying, why is he interested in, in orbital, um, computing hosts and models in, in space? It's, it's the fact that now the build out of all the AI data centers, there's a lot of opposition from people, um, in terms of the, the, the build out and because of power, power usage, right? But if you're in space and if you have solar panels in space, you have, you have no clouds, right? So you've got pretty good access to, to, to sunlight, sunlight to power it, but you're also in space where it's much cooler. And a lot of the, uh, data, the, the data center build out, they have to build with cooling in mind for the, for the GPUs. So he, I think is positioning himself for the future where as a space company, he's already deploying the satellites, why not host, um, the, the compute and the infrastructure of the future in space where he doesn't have to deal with, with the issues of the build out that we're having on the ground. Exactly. And I think it's important because if you try to focus on, on both of them being the infrastructure and the technology layer, it's very hard to do two very difficult things. Uh, and I think Antropic could get an advantage by allowing, you know, SpaceX to build out the infrastructure and just focus on building the, the, the right technology. Whilst if OpenAI is trying to do both, that's where things get maybe a bit more difficult. And I think I would compare this to Nvidia spending much more of their effort on designing the best chips rather than being TSMC and producing the wafers. Yeah. Yeah. And yeah, indeed he's just, he's looking into the future, right? Like if you listen podcast from months ago, he's been saying this, he's been planning this and now he just has the perfect partner who is going to pay for it. Right. Uh, that, that partner that grew their revenue, what is it? 90 X over the last year alone, um, is now going to pay for Elon's infrastructure buildouts. Um, well played Elon. Good job. Friend of the pod. You should come on the pod, Elon. Hey, why not? Um, well, what's, what's, what's, what's, I kind of liked this one as well. Uh, be entropic. You need compute. You sign a hundred dollar billion, a hundred billion dollar deal, um, with AWS for five gigawatts. Then Amazon writes you an equity check back. Um, so Amazon has invested here. Boom. Sign at $200 billion five years with Google for five gigawatts and get a $40 billion equity check. Boom. Um, you got more equity on the balance sheet, signed 30 billion with Microsoft for Azure capacity and Nvidia writes you another check and then Microsoft writes you another check and then fluid stack writes you a check. And then space X gives you GPUs and more megawatts. And then core weave gives you more, more capacity total commitment, total committed spend 380 billion total committed power 11.3 gigawatts. Ben, can you put this 11.3 number in perspective somehow? Uh, I'd have to look it up, but it's a lot of, it's a lot of, uh, power. Right. And, um, I think, I, I think I read somewhere that 0.1 terawatts is where kind of the U S power is, right? So a hundred gigawatts is where we're at as a, as a country. I think it, when it comes to data centers, they've just added 10% of the, you know, kind of country's, uh, data center capacity and in compute. Yeah. And it's interesting to see all these deals where it's not only like, Hey, we'll buy from you guys, but it's, it's almost like, uh, the right to buy in, uh, for Google, for Amazon, for fluid stack is like also equity, right? So they're, they're making purchases in compute at perhaps a fairly favorable, uh, valuation. I, I don't, I don't know exactly, but, um, on top of that, Google is getting $40 billion in equity into Anthropic. Amazon is also taking inequity, um, with Anthropic. So everyone wants a cut of this and, um, yeah, it's, it's crazy how, how, how tangled. Yeah. Yeah. It's awesome to have a, such a network of like the biggest companies are all rooting for you, which is impressive, right? Like if you think about opening eyes, Sam Altman, like there was even some messages that got leaked again from the times when he got kicked out a CEO and then after the weekend he was back, but like, he doesn't like he, everyone wants to see Anthropic win now because everyone has stake in this company. Whereas open AI is basically doing the battle on its own. Um, they don't have such a network of, um, of amazing partners. And I saw this, this photo on Instagram, which basically explains it very well. Like it, it visualizes it all pretty good. One sec. So that's basically currently what, who owns Anthropic, right? Um, if you look at this, so the founders own 47%, then you have Amazon owning 17%, Google 15%, VC institutions, 14%, and then Atom Capital. I don't know what that is. Um, 8%. I also heard for example, that, um, Zoom invested some money in Anthropic many, many, long, long time ago and owned about 1% of Anthropic. But if you look at the Zoom market cap, like quite a lot of that portion of the market cap is actually Anthropic value. So maybe Zoom is a buy now, especially with potential virus coming up. I don't know. Um, not financial advice, but like, it's very interesting to see how all these different companies have a stake in Anthropic and yeah, that means Anthropic will provide for them and then they will also root for Anthropic. It's, um, yeah, there's business strategies actually. I, Ben, you said Darius not a business guy, but, uh, don't you think he played well here? I think he played pretty well. Um, I, I mean, look, he, he's struck up a lot of deals. Um, yeah, I don't have too much more. Rick. Next up we stay in the Cloud universe. Um, and I think, I personally think this is actually a consequence of this partnership with SpaceX because now since Anthropic suddenly has compute, leftover compute, they can actually improve their product even more by putting that compute to work. So, um, Claude announced dreaming in Claude managed agents, which basically means that your agents and Claude are automatically updating their memories based on your conversations you had. And this is something that we've already seen in the agent space where OpenClaw had some sort of memory, then Hermes has a better memory, but Claude is just learning from all of these things and is implementing this themselves. Um, what, what do we think here? This is, this is awesome development for, for the Claude app, right? Ben, I think you, yeah, you first. Yeah. Um, well, so, so code with Claude 2026 happened this week and what was interesting about it was, it was almost entirely like platform announcements instead of actual model improvements, right? So dreaming outcomes, this multi-agent, there was a web hook announcement and then routines. I think they talked about that, but like none of it was, we made the model smarter. It was all about infrastructure, running models in production and basically moving towards this agentic, uh, world. And so that kind of tells you where I think Anthropic thinks the mode is. I think Sam Altman has also been talking about this for years, but you know, Claude is already probably one of the best models. I think we've been saying that although GPT 5.5 has definitely made a lot of strides and it's pretty close to it, uh, models are kind of commoditizing to, to an extent. And Anthropic is now saying, let's build this entire ecosystem, this infrastructure so that you can run your agents on Claude, right? On Anthropic. Um, so that was kind of the big takeaway from here is like between this multi-agent orchestration, the dreaming concept of, you know, that your, your, your, your whole age agentic workflow is just getting better while your sleeping, right? Which people are pretty fascinated with. Um, that was the big takeaway from this whole code, um, Claude, uh, announcement. Yeah. And it kind of confirms our thesis from last week as well, where the abstraction layer becomes more and more important. The, the place where people actually interact with the models compared to, uh, abstraction and orchestration and harnesses is becoming more and more important and have the ability to now outperform the base models, right? If you, if you orchestrate better, then you're going to get better outcomes than if you just give one huge prompt to a frontier model. Um, Alan, what's, what's your experience here? Would you rather have a well-orchestrated swarm of agents or one frontier model? Yeah, I will, I will, I will answer that. Um, before I go there, I want to reiterate the idea of the dreaming part, which I, I very much like. What, what's interesting to me is a lot of the guys who worked in like all the innovations with, with these language models in, for example, the guy who's, who spearheaded, um, Google DeepMind, Demis Hassabis, he, he, um, was a neuroscientist by training, right? And I, I find it so interesting how, how a lot of the artificial intelligence sort of, um, metaphors are all, all psychological in nature. They're related to the brain and humans and in our memories and how do humans consolidate their memory? And we do that by dreaming, right? And replaying certain things in, in our mind, I think. So I kind of liked the naming that they've gone with here with, with, with, with dreaming for, for multi-agent orchestrating. For me, where I'm at with multi-agent versus frontier models. I'm in my view so far, I won't necessarily, if I'm going to build something for users that that is not a very, like that, that is, is a very open and then has a lot of use cases. I would much rather have the, have, have a frontier model and be very much in the loop one at a time. If you're going to multi, if you're going to orchestrate multi-agents at the same time, the drift for these to get in their own way and step on their own toes is much, much bigger. So they can, if you're writing code, they can have conflicts, um, and, and, and managing their work and the state and which ones are finished. And so on, that's a bit more of a com complication. So you might want parallelism by multi-agents at the same time, like the way you build it, there's a lot of patterns you can use, right? You can use one, one agent, which is, is doing a task, you know, planning, operating and doing the task. And then another agent that's validating that. So testing it. Um, and especially when you're using agents, you might want to have a fresh context. So rather than, um, uh, an agent marking its own work, so to say, and using a bias and some cost fallacy, you would rather have a fresh context and, and, and review that. But when I'm actually building software, I'm very much in the loop still. I, I, I would not let like 10 agents at once go off and do a whole lot of work because in my view, I'm still the bottleneck for a lot of stuff, especially if they're complex. Um, I would only do that when I already know the system that I want, and it's a repeatable pattern that the agent can, can go, can go and do. And the scope for the, the ways that I can make mistakes and go off the rails, I already know all those and I can kind of self correct them and, and, and correct for that in, in a scale or a prompt and tweaking the implementation of that. Um, so that's what I'd say for that. Yeah. Yeah. That makes total sense. I, I do think there is a, gonna be a big shift in like people actually realizing this now that you don't need a frontier model, um, or even expensive model. Sonnet is not that expensive, but you can use an open source model that's even 10 times cheaper for easy tasks, like finding something or replacing or indexing your files, right? You don't really need, um, the, the best of the best models for that. But I personally haven't really found a solution for that yet. I'm trying to orchestrate this myself. Um, we're going to talk a bit about orchestration later, but, um, one of the things that Ben submitted as well is the Jeff funds paradox. Where is that one? Um, yeah, Jevon's paradox, um, which basically comes down to inference got a hundred times cheaper this year, but still the compute bill went up. So we're using more tokens like tokens go down, but we also started using more tokens. And if we put some numbers on there, um, that's basically means if tokens get a hundred times cheaper, but we spent 10,000 times more tokens, then our bill goes up by a hundred X. And this is basically playing out right in front of a, one of our face, right? Um, I clearly see this in my usage of tokens. I'm assuming you guys as well, if you're orchestrating agents, if you're using your agent, your Hermes agent, there's quite a lot of thinking. There's quite a lot of distribution of, of intelligence. Um, yeah, this is, it's quite an interesting observation, Ben, what do you think here? Yeah. Um, I think it's such an interesting phenomenon that a lot of people just don't really grasp or understand. They just think, oh, well these models are gonna, you know, get cheaper and then it's going to compress and then it's going to become, become a commodity. And you know, the, the model, the platforms aren't going to make money, but I don't know, think about it this, right? Right. This way. Um, so every price drop will unlock a new economically viable use case. So at $60 per million tokens, you could afford, you could afford legal contract reviews, but then at $5, you can afford support chat bots. Uh, then if you go even kind of further down, right at, at, um, uh, at like five cents, you can have, uh, like all these multi agents that are kind of working in tandem and that they're, you know, be able to build out specific things. Um, so basically like every single, uh, exponential drop in like the price, it just unlocks more and more use cases that are basically unfathomable today. So, you know, we're getting into the world where perhaps every single decision that you're making, uh, whether it's the app that you're pulling up on your phone or, um, you know, every, uh, uh, potential feature has been built out and you can say, yeah, I want that feature. I don't want that feature even before you like stepped into the office during the daytime. So, um, this, this concept is like for sure going to continue because it's continued across tons of other things. Like remember when even streaming Netflix used to be such a pain, it used to be expensive. Right. And then now like you could stream like five different things at the same time. And, um, it unlocked just a new, uh, capability. Yeah. So go ahead, Alan, you had something? Yeah. To build on that as well. I think like when, um, when all these models got so good at coding, I think everyone was shouting, um, to the rooftops that, um, software engineers were going to be completely automated. Um, I think, um, Dario has said this, right. Um, as well that I think entry-level folks will have that problem, but because of Devin's paradox, as you mentioned, because the price of all these, all these things, for example, the cost to produce software has now gone so low, the demand for it also goes so high. Um, so now I think you'll actually see more demand for software engineers because even though, um, it's so easy to produce the software, there's actually more demand for it. You see this even in, um, for example, when they, when they automated, um, getting, getting your money from banks with ATMs, right? The number of bank tellers actually went up, right? Because the cost to run a bank went way down. Everyone else was, was using ATMs to get their money and they could open more banks and provide better services, um, for people at, at the, at the, at the tell, um, much more, much more efficiently. The, the paradox means that even though there's an abundance, um, and, and costs, costs go down because we have an abundance, we'll actually have, have more demand. Yeah. And, and, and each token that's being used or generated or spent is also more capable of doing things. And we are far from everyone using this technology and I'm seeing a future where basically everyone is going to be spending tokens, um, and spending money on tokens as well. Like right now it's idiots, idiots like is like us, we don't know how to code, but we can now build apps, but it's probably like, probably we're spending three to five times more tokens than an actual software engineer would spend when they're using cloud code. That's an assumption I make, um, might be wrong there in terms of numbers, but like it's idiots like us who actually spend quite a lot of tokens. Um, and who are highly benefiting from this paradox playing out in real time here. Ben, you submitted this one as well. So yeah, basically idiots like us, right? Just because idiots like us can build software now doesn't mean software is dead. Um, the SaaS apocalypse has been happening sort of, um, but anyone can bake bread in their home right now yet 99% of us still choose to buy it from someone else. Um, Ben, why don't you introduce one or share your thoughts? Yeah, yeah, yeah. So Farza, um, is first of all, an amazing, uh, creator builder, all of that. And he's so right with this kind of, uh, concept that the software industry is not going to disappear. And Alan, you were just talking about that too, where it's unlocked more capabilities and you guys to be able to build more features. But this baking bread analogy really drove that home for me, I think, and almost kind of rethought, um, how I think about, uh, you know, is software dead and all of that. Um, bread is like really a tiny commodity market. Like not many people are baking bread, even though everyone has the capability to do so. Um, essentially like, you know, yes, it's possible to vibe code your own application, but I think I would agree that most of our friends will not be vibe coding their own applications, their own things for this and that, unless they're really like wanting to be technical and, and, and kind of do that. They might try to vibe code something to like get an understanding of how it works initially. But in the long run, I do think that a lot of people really do want the software to be handed to them. Um, and perhaps they may, you know, kind of try to vibe code something and then it may work a little bit, but then they're like, ah, let me just give this to a professional. The professional will be able to do it. And that's the same when it comes to me, right? I've, I've never baked bread. I go to the baker and I buy my bread from them. Um, and you know, perhaps the bread costs a lot less than it did in the past. But, um, I do think it kind of resonates that, you know, most people will not be building their own software. I don't think that to be the case. Um, they want to go to a professional and they want to kind of stay in their lane of whatever that is. Um, and then the thing that he said, simple products are complex. Like I will always be happy to pay someone to handle the new nuisance. That's just really kind of, uh, echoes like what, what we're talking about with like the kind of baker going to the baker analogy. Yeah. I mean, it's, it's also like the baker has been fine tuning their recipe for years and years and years, and you're paying for experience, right? And the same with a SAS tool that you're using day, they've probably been building and fine tuning this. And they've been talking to thousands of customers to fine tune how this thing works. If you're now going to like think of a, uh, of a Salesforce, for example, right? Like they had thousands and thousands of enterprise customers. They had millions and millions of data points. They've been fine tuning this thing. They've been integrating it everywhere. And if you're now going to try to build your own CRM, um, then yeah, it's probably going to break and it's going to be a pain in the ass and it doesn't have all the integrations and it doesn't have all the knowledge that Salesforce has built in. Yeah. You can say, build me a Salesforce clone, but is that going to be the same? I don't, would you trust it as much as you would you trust Salesforce? I personally wouldn't. So yeah, it makes, makes subtle sense in that, in that perspective as well. One of the, like to stay in the compute side, we just talked about Jevin's paradox. Um, but one of the things I just learned was that Amazon bought a, a chip manufacturer called Annapurna Labs, um, quite some time ago, an Israeli startup. And that startup turns out to be creating chips now. And if, um, Amazon CEO said in their letter that if our chip business was a standalone business, our annual revenue run rate would be 50 billion. And if you think about that and you compare that to other chip manufacturers, that would basically be, I think the second largest chip manufacturer in the world. Um, but because of us export, um, rules, they're not allowed to export these chips. Um, so they can only create this for internal usage. Ben, I know you have a history at Amazon. Were you there when Annapurna Labs got acquired and do you have some, some background information on this maybe? You're muted. Yeah. I don't have any background information on this. Um, and, uh, yeah, it seems like every company is kind of trying to play their hand in, uh, in this whole like chip business. Um, and it's interesting to me that, you know, this indicates that Amazon has $50 billion worth to kind of just spend even internally. Like, I wonder if they've been doing this to build out their own models. Um, I know they haven't really come out with a model that at least has, um, yeah, go ahead, Rick, you were going to say something. Um, I think Amazon is amazing investing in the future and they invested in this Annapurna Labs. I don't know exactly when, maybe let's ask Grok. In 2015. In 2015. So yeah, 11 years ago, Amazon invested in this Labs, um, for the future potential of chips, right? And now they have basically a whole chip business under themselves, but because of the regulations, they cannot really monetize it to their potential. What I found impressive as well, other news that actually happened today is, um, this week is that Amazon basically open source their supply chain services, which is another, like it's huge news, but that's another testament to Amazon investing into the future. This is not something that just happens. They are there like that just happens, right? This is something that they started building 10 years ago, 15 years ago. And now the supply chain is so robust and they have like, what, what, what did they announce recently? One hour deliveries in the U S like now everyone can actually make use of that delivery system, which they've been investing in for years. And it now it's going to pay off. So like, it's just impressive to see how Amazon kind of just like Elon thinks 10 years ahead, invests in things that then come to fruition and that then become actually the big revenue driver. Same with AWS, right? Probably was a cost in the beginning, build it for themselves, but build it so good that now basically the whole internet runs on it. Um, so yeah, it's just like kudos to Amazon and, um, Ben, you still no regrets for leaving that company? No, I don't have regrets cause I wanted to build my own thing. Right. And, uh, it was a great learning experience, but, um, but, and it's an amazing company and this side of the business was closer to where I was when I was there. Um, but, but yeah, like, like I I'm, I'm curious to understand. I need to dive deep into like why they wanted to kind of open source this whole, this whole area of their business and like what the, what the kind of motives and, uh, motives behind that, right. I mean, where I think it makes total sense, right? They have the network, they have the careers running around in every city for every hour delivering door to door. Um, and it is an industry that had to be disrupted. If I want to send some, if I want to send something with old school post carriers, it's going to cost me a lot of money because they don't really have that network effect. They don't really, they're not driving around anyway. Like, right. They actually got to go to places. Amazon is doing it anyway. So what's the difference for adding three more packages from a external supplier, uh, from a third party? Like it's, it's a revenue generator because they're driving around anyway and they're paying their, their careers anyway. Right. Um, so I think this is great. Um, and I, I didn't really check the stock prices of like FedEx and others. Um, but I know in Europe, it's very pricey to send some stuff around. Like we literally use hacks like, Hey, um, buy this on vintage from me. So it only costs five euros, uh, transportation instead of 20, right. That's, that's literally the solutions that we have to come up with in Europe. Um, and yeah, I think this is just going to disrupt another industry, which yeah, kudos to Amazon internet, um, internet businesses turning into the real world. Anyone else had a, had something on that, by the way? No. Just very smart move to, from them. It seems like they've, they have a good holding on Anthropic and taking the play of providing compute in the future seems like a solid bet, especially when you consider Jeven's paradox is going to continue growing that demand and, uh, selling the tools, the tokens is the best way to go. Um, it's, I will, if we go back to the Baker analogy, I would say it's like selling the flower instead of the finished product, right? Well, dive a bit deeper into agents, um, because the race of the agents is continuing where mold pot, open claw, claw, claw, claw, whatever. Um, a few months ago was the big thing. Um, now there is a bunch of other agents either integrated in the big players products or you have no now's research, news research. Um, I think that's a lab in Canada if I'm correct, but I'm not sure. Um, um, they came out with Hermes agent. And this is also why Alan actually, why we invited Alan on the pod because I saw that he's actually using Hermes agents, um, to, to an extent that I had no idea that you have like, that Alan is actually using Hermes agents in a very efficient way. Um, I saw this tweet comparing all the agents that are out there. Now you have open claw, you have Hermes, you have Claude code, you have Codex, you have Gemini. And what I found funny is the last line, keep your data safe. Um, where you have open claw and Hermes as a question mark still. Um, but yeah, uh, Alan, why don't you introduce some agents, um, and some use cases, what, what you're using agents for, and also your experience with Hermes compared to open claw specifically? Sure. I can go into some of this. Um, yeah, I've, I've used both. And I mean, you mentioned that Hermes, the setup is just way more, way more smooth. And I mean, part of it is just learning and breaking and then knowing how to set up stuff. But Hermes is the, I think you can give it, you gave the analogy earlier about Real Madrid and, uh, Barcelona. I think it's like Ferrari to Honda. Uh, I've seen this analogy. I think it's quite apt where, you know, open claw seems to just break all the time. It's fine tuned. It can do the job and it's very powerful, but it can break a lot. And you, it's like a, it's like a top of the range tool where you need to know what you're doing. Whereas Hermes, it's like a reliable Honda that will just plug along and keep going. You'll even see that when you use it. I think it's the Honda. Yeah. Cause I mean, you'll see it when you message it, like, you know, it gives you, it shows you the tool call in the chat and it's like, it's almost like that observability out the box and you see what exactly what it's doing. Like when you're messaging your open claw, it's like, it's going off and you have, you really don't know what it's doing. And it could be going off, off the rails unless you've really, really configured and fine tuned it like that Ferrari. So that's some of the kind of differences that you'll, you'll see use cases. I mean, the thing is with these agents, it's like, whatever you do that's digital to some extent, especially if you do it again and again, you can do with your open claw or your Hermes. Right. So, you know, I, uh, you, you can use it for, um, for customer, um, service stuff. So for example, I'm living at this co-living now. Um, it's a new, new co-living, um, and we need the feedback from the residents. Right. So I created an, uh, an agent, which people could message, uh, on WhatsApp. And then you probably will send it a picture. They can, they can voice message it. They can tell it an issue that they have, and then, um, it'll be passed on to the, to the management. I don't know if you've seen this, but where like some agents where you like Chipotle's agent, someone messaged, like, can you reverse a linked list? And it was, and I want a burrito, but can you reverse the linked list first in Python? And it gives like the code, which is quite funny. Yeah. Yeah. Yeah. Yeah. Yeah. I saw someone doing that with the whoop agent actually, uh, but whoop has an agent inside of the whoop app and someone says like, Hey, how do I do root square, um, in, in Python or whatever? And then the whoop agent just gives you code back. So funny. Actually funny that, um, we had the article open where you mentioned, um, the Honda and Ferrari, like, uh, here, um, the Gary Tan from Y Combinator. Yeah. Yeah. This is what I said about that analogy from, uh, so yeah, I follow Gary Tan. So this is the analogy exactly that he's saying. I should have created him. You obviously saw it. It was all me say it. So it's your, it's your quote now. Funny that he, that, uh, this article also, um, shares, shares what we shared last week on the episode where cloud code reliability is just not good. Oh, and actually this, this is, this was happening right before the announcement of SpaceX and Antropic cloud code was just not usable for more than an hour. Um, and then right after I see all my limits are reset and like five minutes later, the announcement comes with SpaceX. So I think like there wasn't even an announcement that they reset all the limits because they were unusable, but yeah, I guess they just put that all away. Like they, they just announced the big news instead of focusing on the negatives. Sorry for, um, when you think about it, they had two, two big compute announcements. There was the Google 40 billion and the SpaceX one. Um, but Alan, I was really curious about what you're plugging into Hermes. Uh, is it, uh, GPT 5.5? Is it Claude? You mentioned Kimmy as well earlier today. Um, what's, what's, what do you have and what are you using them for? Like which ones are you assigning to which tasks? Yeah. Personally, just because, um, OpenAI's OAuth is free. I just use, um, um, GPT 5.5 and I keep it simple like that. You know, I've used providers where, you know, it's for other fallbacks where it's cost per token, but you know, if I get a free, like frontier model, then why am I not going to, why wouldn't I use it? So I, I just use the best model that I can. And I find, you know, the best models can handle more context. They are faster and they, they are more reliable at tool calling as well. Although I have found that with something like Hermes, much more than OpenClaw, if you do put a free model in there, Hermes is much better at getting it to, to, to do tool calling right at the box and, and, and sort of trundle your way like that Honda to, to, to the end goal. And you also mentioned translations that you're using Kimi for that or, or something else or GPT 5.5. Just to be honest, I think most models are pretty good at translations. I'm just using, yeah, I'm using, I'm using, I'm using the GPT 5.5 for that. Although I've used some of the free models and I've used Kimi as well by other providers. Then, you know, the Chinese models of course they're built by the Chinese. So it's pretty good. If I'm doing my language learning in Mandarin, then I can also use that. Honestly, GPT 5.5 is good. It's free with the OS. So I would tell anyone to use that. Yeah. I think long term, this is the biggest risk of companies like Anthropic is these open source models that are actually competitive and being hosted on either people's local devices or people spinning up their own cloud GPUs. So I think that's where then companies like SpaceX that have the infra, have a much more secure investment. But that's, that's, that's just what my personal belief, just seeing how fast all of the models are catching up with each other. And also realizing that whilst I'm working today, very often, I'm not even using the frontier model like Opus 4.7 or, or GPT 5.5, because I want an answer faster. And I'm not asking it something that's that complicated. Yeah, exactly. Yeah, I agree. One thing what I realized as well, yeah, no, to build on that as well, I actually let my cloud go and do some deep research of what models I should plug into my Hermes agent for different use cases. And one thing that I learned was that there is this thing called time to first token. And I didn't know that existed. But for that reason, it actually said, now you should not use Grok for these use use cases. It's better to, for example, use a Kimmy or use a deep seek because they have a much shorter time to first token. So that was something totally new for me that I learned. And yeah, it was quite interesting that with the research it came back to. And the other thing that that's been circling on my timeline about regarding Hermes agent or agent specifically is this sharing the screen. Introducing the printing press, a CLI factory and a CLI library built with Treven. Most APIs suck for agent, most MCPs suck for agents, most official CLIs suck for agents. They waste tokens and time. Peter Steinberger started making his own because of this. And that's basically why we saw that tweet where Peter Steinberger spends how many tokens? 50 million tokens in a minute while I only spent 25 million tokens a month. But here, this video is actually explaining it. I'll start from scratch. I saw that tweet of Peter and I was like, what are the business opportunities here? Because he launched like 10 different services, all targeting agents with CLI. And I'm like, okay, there's something here. And for that reason, like you saw the example here where if you just ask your agent to play a next song on Spotify, it would cost you 47,000 tokens. Like I can almost build an app with 47,000 tokens, right? So agents are just eating tokens. And I think this is basically a solution for that or a solution, at least a way more efficient way of using tokens. I'm not going to play it fully, but... And Peter released, I think he released all of his solutions free. So anyone can just send the CLI calls. But if you're saving an agent that much money, then there's a, you can make a small arbitrage and still be a useful tool. Yeah. Yeah. And like this, this thing, like the printing press.dev basically combines different tools or different information sources into one. So like, for example, if you want to use your agent to book a flight, right, it can do that already. And it's probably gonna eat quite a lot of tokens, but then there could be some other influences that impact your flight booking. Like for example, your favorite football team just played and they actually won. So they're going to the next round. So then you like, um, then you can like, your agent doesn't really know that now it could know these things stupid example. But like, if I scroll through the catalog here, it's pretty impressive. Um, like for example, uh, TikTok shop, Yahoo finance, a Docker hub company go that like all different skills that you could just plug into your agents. Um, where this, the previous APIs not really allowing, um, good use, um, or not really actually allowing these use cases. And now with this whole, what is it? A repo. Um, you can just, I plugged, I just asked my cloud code, scan this, see what's, um, what's best for my use cases and then plug it into my Hermes. And it came back with a list of six or seven that were actually good for me, plugged it into my Hermes. Now I can just use it like, like, beautiful. Right. Um, Alan, have you experienced with, um, experimented with this yet? I haven't played with printing press.dev, but I'm definitely going to give it a look. I should say that. Yeah. Peter, Peter Steinberg, a lot of his stuff I'm using. So obviously I'm using open claw, but for example, his bird CLI tool, which plugs into X, which allows you to access that. Um, that's the main thing I think with the agents, right? Is that in order to make it really useful, you have to connect it with a whole load of things. And if you don't have connections, then you can't really get it to do what you want it to do. So part of the reason why it's such a, it's kind of difficult to build agents is, you know, you'll see all the different examples here, but you are the reason why it's so difficult to build is it's kind of like the wild west right now at open source. There's so many different ways to plug into your agents to connect, you know, whether it's CLI APIs, MCPs. I think one of the things that I'm interested in, and I think, well, might get some traction is the idea of the new UX being the chat. And because the new, the new UX is the chat, you kind of limited by the text that you get back, right? Or, or the image. And so I've seen one of the things I liked that Anthropic have done is with MCP, right? That they created. They, I I've also seen as well, MCP front ends as well, where you would have any app, like maybe it's your booking.com. You can connect it to their MCP and suddenly you can have a chat with your agent and then it could go and book you, your, your, your, your, your hotel. However, it will show you that a page with your, the bookings, um, given what it knows about you, that, you know, you like like city breaks, and then you can go and click within the chat for whatever thing you want to book. So I think there's, it's still early and there still needs to be a much more improvements in terms of the UX where everything is kind of through a chat and people are connecting, but right now it's just text and talking to a bot where you still need that user experience to really be improved. What you see here is our docket, um, that we have for the podcast preparation and we submit our, our ideas, um, or which are these tweets that we're looking at, um, via telegram. And then they get into a database and via MCP, Claude can read it and Claude built a custom dashboard for me. And now I can literally chat about the topics with Claude and generate an introduction, generate an outline, like MCP really changed the game more and more with all the apps that I'm building and thinking about, like, I'm thinking, how can I actually make this even easier to access? Because I don't want to go to five different apps. If I can do everything from one chat window with, with MCP, it's just such a better, so, so much better user experience. And it's also interesting what you say there, um, about the future might not be apps. It might be chats. Um, that's, that's in line with like last episode, we talked about the end of Apple potentially, or the end of the app store, at least if everything becomes agentic. Um, yeah, that's, that's a future that I definitely see happen. Um, which I'm not sure if it's exciting or not. I'm not sure what I should think about it. Ben, are you excited about an agentic future or are you kind of, um, scared that they're going to take away our app building, um, like the opportunities and building apps? Um, I think there's a lot of opportunity here, right? Um, we were talking last episode, even about just being able to sell to the agents, right? Um, I, I also though, like, I do wonder whether chat is the right interface, um, on one end, right? If you think about where most of people spend their time when they're at work, it's usually on Slack and they're chatting. They're chatting with other employees. In this case, you're just chatting with a chat bot. That's like the Slack of the Slack equivalent for, you know, chatting with the AI has just been chat GPT or Claude up until now. But is that really the right workflow or is it more kind of like Claude and Codex are kind of going in this direction of like, you're like listing out all of these like things that you want to be done and they're all kind of working and then you're like checking on them and seeing how they're, they're progressing. Um, yeah, no, I mean to kind of like tie this up, like, I think nobody really knows like where, uh, this is going to go, but the, the, the, the chat interface to me feels like the first iteration of this, there has to be a, like a better way to be able to almost like a mission control area, right? Where you have everything that's going on and you can see it, uh, playing out in front of you. And then you can like double tap in and then, you know, you could basically have the 10,000 foot view, but then you can drill down as, as you want into these different workflows and then drill out as you'd like as well. Um, so I don't know. It's, it's, it's hard to say. I, I really do think the more, more that we've gone through this path, like I do think that SAS is not as dead as people think it is. Um, I think that the prices are going to go down and you know, they'll be compressed, but people really do want to pay for an experience. Um, and they want to be able to see this kind of, uh, UI that accomplishes something, whether it's booking flights for me or whether it's, you know, helping me with my, uh, you know, chores for the week, something like that. To me, what we're seeing with printing press, it looks like the new WordPress and each one of these tools is, uh, can be considered an app or a plugin. And it's an exciting time. I think if, if you look at these as opportunities to be the middleman between connecting the agents with the real world, I think you, there are opportunities all over the world to kind of localize these CLI services, whatever it is, because I mean, for over here, you're seeing some things that are as simple as dominoes, you know, food and dining. And I believe that in the next few years, everyone's going to be ordering their pizza, not through a mobile app, but just by asking their device, Hey, order me a dominoes pizza from the closest place. And if you're able to build that, that infrastructure, the plumbing that connects the real world to the agents of people, I think that that's the, there's multiple million dollar opportunities there. Mm-hmm. Yeah. And if an agent can do it and you can put it into a front end and front ends are not dead yet, people are still used to using front ends. They're just not using, they're not just, they're just not used to using front ends that actually get shit done. So if you can take one of these things, um, connect it and put a front end on it, I think there's still an opportunity for the next one, two, three years to actually build something useful that can serve people that wasn't possible before. So that's, that's a very good point. They built a the plumbing and now we, we idiots, Vibecoating can go play with it and send it into the world. Yeah. And then I also think there's this concept of like today you ship SaaS as a finished product that's the same for all your users, where the future might be that everyone has the same starting point, but the app is slightly adjusted according to how the user's preference is. Mm-hmm . Um, so it's almost like SaaS can be adaptable on an individual basis. That's, that's actually one of the... I mean, that's what OpenClo and Hermes is in a sense, but yeah, I think we'll see more of it from, uh, from, from different places. That's one of the best practices in design, um, in the 2026 design guidelines now as well, is generative design. Basically that your design like shows the user that it actually looks like it's generated for you and it feels personalized. It's not super personalized yet, but I think we're going to get to a point where actually websites will be personalized for the user that visit that website. But we had this in 2019 where based on user preferences, cookies, and what we knew about a user visiting a website, you would show a different website. So it's been around, but now with AI, you can actually like, you can get this to a next level and you can even generate images or landing pages based of the user that's visiting and based of the data that you have on, on the user. So yeah, super exciting times and opportunities. Ben, what? Yeah. And I, I, I was, I was just going to say anecdotally with my business, with the sales product, I've especially leaned into this where, uh, sales echo can be used by anyone from a founder to, uh, you know, a real estate sales rep to someone in the insurance industry. And all of them have very different sales problems and kind of sales journeys throughout. So, um, by having them just answer a couple of different questions when they first like log onto the platform, I'm able to hit them with like 10 different email marketing, uh, 10 different emails, like in the, in the marketing flow that are specific for that industry. And then I'm able to like change the components of like the kind of page. So instead of it saying like, just start sales call, it'll say like, you know, how to start, uh, like this real estate call. Right. So I'm really leaning into this because like, uh, cloud code and all these, uh, AI agents are so good at being able to record all of this data about your users and then be able to think about, Oh, well, these are your top five ICPs. Why don't you build this landing page for that person? Or, um, and the landing pages are hard because I don't know about them when they've come to the site, but once they've created the account and once they've like filled out a few details, I'm able to extrapolate so much about the user. And I basically build off of previous users to understand like, okay, what do we think this new user will be? Will they be in the real estate world? Will they be in the insurance world or will they be a founder? Things like that. So I think we're going basically from like, uh, everyone sees the same user experience to industry or ICP specific, uh, apps to then, you know, Rick, what you're alluding to in the future where every single person is going to have their application that, uh, perfectly aligns with like how they would want to go through the user experience. So guys, um, fireside round of updates. There's just too much, too much news in AI land to be able to dive too deep into it. And the first news that we're going to comment on is drum roll. SpaceX AI is announcing Grok build. Um, it's not announced yet. It's not official yet, but it's coming soon. It's basically the cloud code or the codex for by SpaceX AI. Um, again, the, the, um, the abstraction layer, the app layer, everyone is building it. Big labs are building it. Um, I'm wondering who they are going to target or what is this going to be for, or this is going to be the new cursor. How is it going to be? Um, Ben, any thoughts on Grok build? I wonder how long they've been working on this because cursor and that whole kind of acquisition deal probably, uh, through, uh, a wrench in kind of that whole plan, or they probably had this in the works before a few weeks ago, I'd imagine. But at the end of the day, like, you know, Grok build is only going to be good enough if, if the models that support, you know, that, that are supported by it. And, um, my assumption is that Grok build will only work with like Grok 4.2 or whatever. And to my understanding, uh, Grok is not that great when it comes to coding, right? Um, so this is probably a good groundwork for perhaps the future of, of whatever Grok's models are and, and if they are good at coding, but to me, I, I, if I were betting on SpaceX's kind of coding, um, roadmap, I would say that the cursor aspect is way more, uh, exciting to me than, than this Grok build feature. I would say that the only value I could see coming out of something like this for Grok is that they are building their own harness for their own model. And maybe that helps them to maximize how effective it can be used, but its potential and being adopted as super hard to, to believe in it, given the competitive space it's entering. Yeah. Elon has said that six months behind in terms of coding. So I don't know, I might wait. Next up another XAI, your customer support needs a voice agent built for the real world. So XAI is now also introducing a Jarvis-like voice agent. Grok voice thinkfast 1.0 handles complex workflows with speed and accuracy. Um, any, any thoughts here? Who's going to implement it? Ben, you and your sales echo product, maybe this is useful, no? Maybe it needs to come down to price. I did, I did see that open AI just came out with their GPT-5 level, um, real-time voice agent. And this is just another anecdotal kind of interesting thing. Um, that's a little technical, but I'll just get into it. Um, so basically GPT-4.0 was the only model that open AI built and trained it on a voice to voice level, right? So, um, all of the voice AIs that we've been running on like the five series, what they're doing is they're going speech to text and then text to speech. But instead up until this point, uh, before, since 4.0, like that was the only model where they were training voice. So how do you train voice? Every single sound that you're making, they're predicting the next sound. It's way harder to train that way. Um, and then GPT-5, this, or the, the, the real-time two is using the GPT-5 class. So this is the next version of it. I'm way more excited about what the potential is of this because the 4.0, uh, model, when that came out with the, with the voice, it was amazing, right? You could, you had it and it was like able to, you'd have, you were able to have a real conversation. It sounded like he or she was able to even like intonate like a human would, and she could speak fast or slow and all of that. And so I've, I played around a little bit and it seemed pretty impressive. Um, just like, you know, kind of like using the app and, and kind of doing that. Um, but this is way more exciting to me than whatever Glock is, is working on their voice. Cause they're, they're basically six months behind. Yeah. Luca. Luca. I, funnily enough, use the Glock voice quite a bit. Like when I'm driving, I just hit the voice and I ask it for like just some searches or some basic questions. And I, I've been enjoying using it. I can't compare it to anything else so far, but what I, what it's, what seems to work about it is whilst you are speaking, it's already thinking. Sometimes if you take too much of a small pause, it actually interrupts you and starts answering. But then if you start speaking, it will stop and continue listening. And it's been useful. I've been finding it, um, good enough to work with. And, um, yeah, I, I think these, seeing these integrated into, into more areas of my life would be useful. Yeah. Um, I'm, I've been loving like whisper flow or, um, super whisper as like, just for like voice to text, not really having text speaking back, but that improves my life already so much. And then indeed when on a walk, if these, these models, if I could, one thing, why I'm not actually using it on a day to day basis, the listening to an AI model is because they just talk like normal people. If I could 2x their speed, um, of output, then I think it would be, would be more useful for me to use. That's how I listen to podcasts as well, 1.5 or 2x. And then like, I'm so, I'm still faster reader, right? So it's, I can just as well read while walking, but if they talk 2x, then I think it makes more sense for me to, um, to use these. So yeah, in a sense, uh, maybe that's an app to create, um, a 2x, uh, voice agent that speaks 2x fast. Next up, um, cursor introduced orchestra slash orchestrate is still the skill that recursively spawns agents to tackle your most ambitious task with the cursor as a K. We continue, um, on the cursor here where cursor is actually building out the best harness, um, I think in class and potentially like is going to take over. Um, we're not sure yet. Codex is great. Cloud code is great, but cursor is doing amazing things. Um, Alan, have you been a cursor user? I used it very early on. Um, I was on the waiting list for, for cursor, like, you know, 2023 or like 20 very early in the first few million people that used it. Um, but since then I moved over to the, to the coding agents. I just, I'm, I'm more of a hacker. I like being in the CLI, like typing in the terminal. So that's more my style. But then you don't really see the agents like you see here on the screen, right? How it gets orchestrated or do you, you're not really a visual person then you'd much rather see the text on the terminal. Not really. There are different tools that I know you, that you can use for, for the orchestration where you can see them. And, but I, I have it old school with just a few different terminals when I'm building. That's the way I do it. Yeah. That's the interesting difference between a marketer like me and a real core deaf like, uh, like, uh, I, I prefer to see things like this, how it actually works. Um, and I rather don't touch the terminal. Like that's what the first thing I wanted to set up when I was trying to use OpenClaw was to be able to see how it actually is orchestrating and what it, what it actually is doing. Um, so yeah, that, that's funny. The difference here, right? Next up, OpenAI again. Um, Codex now works directly in Chrome and I've read that this is actually a plugin that's great. And the browser use and the agents being able to use the browser is finally on point for having good enough speed that it is useful beforehand. Like browser use was slow, was buggy, but apparently now it's on point. It's great. Um, I have not tried it out yet. Oh, actually I think I tried it out without knowing it because yesterday suddenly, um, my codex started making screenshots in, in, um, in, in, in Google Chrome when I, when I was refining a website. So I was like, huh? How is it doing this? I guess it's this, um, codex in Chrome. Awesome. Um, agents are extending their skills. So great news. Dan, anyone else? Yeah. The Hermes agent, we, we already clicked, we already double clicked there. Right. Um, but this video basically shows what a Hermes agent is capable of and what is the new stuff that they shipped. What I found most exciting here is the, um, the built-in agent Kanban board. And this comes back to what I just said, Alan, um, seeing what an agent is actually working on and what it's doing. And, and that's why I think that Hermes is actually a great choice for anyone. You don't have to be a dev. You can go and use it and it's built in like these things are built in. Awesome. Um, I'm, I'm going to continue explore this because I see, I finally see something useful for me where I don't have to continuously fix and optimize. Um, but they are actually doing this for me. So awesome. And thanks for also sharing all your use cases, Alan, anything more on the Hermes agents, because this is the last agent slide we have. I think they have a cool name. I like the name Hermes. Yeah. Yeah. Yeah. Agreed. Yeah. Hermes, I just keep building with it, to be honest. It's just like you see where I go with it. Uh, I think going forward is make, I think there's definitely something in building is building Hermes, um, or these agents where it's standardized and people like who are non-technical or don't want to invest the time building can just one click and, and use it and connect everything they, they, they want out the box. I think that is a big gap at the moment. I don't think anyone's captured that. And the memory, right? The memory is a differentiator. When I was explaining to a friend yesterday, what the Hermes agent actually does and, and they gave the example of if you could talk to chat GPT in a thread and you start a neck thread, like the memory is very limited and it doesn't talk to all your information. Hermes agents, you plug into your, your files or your GitHub or your report, whatever you're syncing. And it's like, it's, it just automatically updates the memory. Like, like we said about dreaming and clawed now as well. Right? So yeah. Um, I love to see these things progress as they are. Next up is Apple. Apple has been behind and Ben said last week that Apple only has 12 months before becoming irrelevant if they don't move into AI. Now Apple iOS 27 will let users choose third party apps or AI models, including Claude and Gemini to power features across its devices. What's funny here is that I don't see open AI. Where is open AI? Sam Altman, um, loves Apple. It's like, um, he, he tries to copy Apple everywhere, but then, um, they exclude him in the models being used in the iPhone. Why is that Luca? Well, he did hire Johnny Ive from Apple to make his own killer device. Everyone hates Sam and everyone is included. This is the thing with Sam. I think he's, he's extremely ambitious. He's, he's started out as the leader and everything. He wanted to take over every single vertical that AI was introducing to this new, uh, new era. And maybe it's spread him too thin, burnt a couple of bridges along the way, and we're seeing it in the partnerships that are happening. Hmm. Yeah. He's maybe, yeah, he did indeed maybe throw in his own glass or his own door. I, I, yeah, I, I also do think that, um, Anthropic and Apple are more aligned when it comes to privacy, right? So, um, you know, Google and Apple, Apple and Google have this weird partnership throughout the years where there's a lot of, like, I think Google pays Apple something like $10 billion a year to be the, the default search within Safari on the phone. Um, and so that's probably where the Google and Gemini thing comes into play. But then with Claude, you know, Anthropic is much more privacy focused, um, somewhat to a fault, maybe some would argue when it comes to certain things, but then when it comes to this, I think they actually, um, are more aligned when it comes to Apple. Whereas open AI is full capitalism fully like, no, if you guys use our products, we may serve ads to you. We may train on your data. We may do all of these things like, you know, so, uh, open it. I mean, I've, I've used Siri on, on Siri and chat GPD has been integrated for over a year now. And as I just said it, my phone just went off. Um, but, uh, but basically, uh, you know, like, like it doesn't work. It just doesn't work. The integration does not work at all. So I think this is them saying, okay, we're going to cut our losses. Open AI is going to do their own thing. And then Apple is going to partner with someone else. Next up is Mark Liu. And basically I started a challenge on my YouTube channel inspired by Mark. Uh, my first video day zero of my 30 day challenge, um, is featuring Mark Liu, not really with him, but, um, me taking away some cool quotes that he shared in another video with a starter story. Anyway, his 2026 goal is make all startups AI first. And he's starting with data fast, a product that both Luca and Ben are using. I didn't submit this. So Ben has a, I think you did, right? Is there a reason for that? Yeah. Yeah. Yeah. I submitted it. And the reason why I submitted it was because of the last, uh, the last line, which is AI assistance will do the UI. So we've been talking a lot about on this episode about front ends and what the importance of front ends will look like. And basically Mark Liu is saying that front ends actually mean less, right? So we're all theorizing, you know, should the front end change to be more dynamic? Should the front end be, uh, like a, like a mission control kind of for your, all your apps and all your workflows. Well, Mark is saying like, no, there may not be a front end in the way that we think about it instead, actually he's leaning super into the CLI. So even this week I use data fast, love that product. Um, there was an announcement that he, uh, created like a, a CLI. So that's actually, even though it shows it's in progress, like I saw that I got that within data fast yesterday, he is trying to make everything, uh, in, in, in the form of where AI agents can use his products as well as possible. Um, he did also just come from Stripe, um, connect. And I, one of the other indie hackers that was there is this guy named Jack Fricks, who I think I told you, Rick, he's building a product called post bridge and Jack for the last month or two has been saying, I'm going to double the revenue of my business by leaning super into AI agents and having my agents be able to having other agents be able to use post bridge in his case, to be able to post on all of their social medias. Um, so I think maybe Mark got a little bit, uh, AI pilled by Jack perhaps at the conference cause they were hanging out a bunch. And then I can see like, you know, whether it's, uh, you know, MCP CLI, the open AI endpoints, all of that. But what I find super interesting also is this generative UI, like what does that look like? I think that is kind of what we were talking about also on this pod where the user experience for various, uh, customers varies. Um, and you know, you basically can now drill down to the ICP level when showing an application to someone. Um, there will be almost this, like, you know, there's a company called revenue cat or a super wall that help build the perfect paywalls. And up until like this point, like people have done a lot of ABC testing when it comes to paywalls, but that's now going to go beyond the paywall to other parts of the application. Hmm. Luca, what do you think? Yeah. On this tweet, what caught my eye was the first two things he mentioned, which is open all API endpoints and LLM.txt and Markdown docs. As you guys might remember last summer, I built a small project, the thunder crawl that was in this niche of LLM.txt. I'm not sure actually, I still believe that that's worth doing, but, uh, it's both of them, including opening the API endpoints, I believe are an SEO play or an AI SEO play for the open API endpoints. What that allows for your website is so that when someone is like searching on chat GPT or, or, or Claude or whatever, they it's much the, these LLMs can call your API endpoints to get the data about your websites. And when you see them thinking, this is more likely for your websites to appear in them thinking. Um, uh, I noticed it with one of my latest projects. And what I know is that usually in standard practice, um, a lot of these API endpoints are closed for security. So people can't take all of your data, but in today's world, you want your websites to be as accessible as possible through an API so that, uh, it is indexed much more easily by these, these AI engines. Yeah. Makes sense. Yeah. That's, um, yeah, the way forward, I think having, having open, having open endpoints, having agents interact with your site and potentially every interaction might become a revenue event, right? If, um, if, if you have valuable data that can be shared with agents somehow and people want to make use of the data that you collected. Yeah. And you would be more likely to be listed as a source when people are asking about your sphere of influence or whatever. Yeah. That's, that's maybe the way to, to hack the agent SEO that Ben is trying to figure out as well. Next we have, Oh, we have the end of, um, of stock prices going up after announcing that people, that they are laying off employees. We've seen a few cases over the past 10 weeks where you had big companies, um, laying off employees and their stock price goes up. Now this week we've seen two examples of stock prices going down after announcing that they're laying off employees. First we had Coinbase. They went down after, um, announcement of that they're cutting 10% of the staff. Maybe they also went down because they didn't hit the revenue targets, but Cloudflare stock crashed 16% after hours, after they announced that they lay, that they're laying off thousand hundred employees. Is this the end of, um, the arbitrage between employees and, um, and stock going up or is this just a coincidence that this happened this week? Any thoughts boys? No, this is just the news then. Okay. Interesting observation. Next we have Google launching, Google generational run. Google actually has almost become, um, almost the most valuable company in the world. It's only a few hundred billion away from Nvidia. A few hundred billion. Funny that I say it as it's, uh, just a little, but, uh, yeah, that's just one day of a green candle, right? But Google just announced, um, that they're launching the Fitbit Air. Um, I'm wearing a Whoop. I've been wearing this for 630 days. I have a hundred days left on my subscription and I say subscription specifically because I pay $260 per year for this thing. Now Google comes out and they, they price it at $99 with no subscription. It's so over for Whoop people say. Is any of you wearing a wearable fitness tracker? Uh, I have the Apple watch. Um, and, but I'm, I'm not as health focused. I'd say Rick, you're probably pretty into that. You've got the Whoop and all. Yeah. Yeah. This, this is going to, I think very much change this whole, um, the whole view of that. Like Whoop has been at the forefront of this, like the no screen trackers. There are some others, there's Aura Ring and then there others really tried. But what, what made Whoop stand out is like the analysis, right? The data and the analysis. But if Google now can offer a similar device with similar tracking and similar, um, analysis, then for me, there's not really a reason to continue paying two and a half times more for a Whoop per year than I would pay for a one-time payment in the, for the Google one. Um, so I'm, I'm thinking the future here of Whoop, either they have to have a very good response to this or Whoop really becomes a status symbol. Um, and yeah, I'm wondering like, um, if they're going to respond to it, I have 110 days to decide if I'm going to extend my subscription or not. Um, so I'm going to wait a bit until there is enough, until there's enough reviews on YouTube before I make the decision, but it's another decision that I'll have to make, um, pretty soon. Yeah. The only, the, the, the last, the one thing I'll say about this is it is smart that Google is, you know, they partnered with Steph Curry. I think there's also a band for $130. Uh, that's like the Steph Curry band. Right. And I think, uh, Google like acquired Fitbit a long time ago, five to 10 years ago, I think, but, um, they, they just had a really bad branding problem with it, where it was kind of like the fitness tracker for, uh, like not so fit people. Right. And so I think this, this version, the Fitbit, the Fitbit air is really going after whoop where, um, it's, it's, it's way sleeker. It almost looks like a whoop. You, you could, you could be looking at it on someone's wrist and you may think, Oh, is that a whoop or not? So it has that kind of like branding identity thing. Um, but, but to your point, Rick, like a lot of people in the fitness world, they love the whoop, not because of the, the hardware aspect, but because of the data, right. And the software and the app and, and how it, you know, just connects with everything and, um, is really able to, uh, like tell you how fit you are and your, your age, right. Your biological age seems like an amazing feature. And I'd love to see what, um, you know, how this works with the Fitbit air. The whole thing was like, um, that all that the data is very closed off and you couldn't really access your data externally. Also it's called Fitbit air. So if I'm thinking about Apple naming, you have the Mac book area of the Mac book pro, maybe the Fitbit pro is going to be announced soon. And is this just a test test case? Who knows? Right. Look out the last one on this topic. Yeah. Just the last line that it's powered with Gemini. That's, that's obviously an advantage compared to whoop. And then obviously a one-time price then makes me think that obviously the users will be a bit of the product and they'll be able to improve their service based on what data they collect. Exactly. The, um, personally like a positive outlook for how health and AI can overlap. So I think that's generally not good. The powered by Gemini. Um, I read that this is, um, included for Gemini pro users, um, or Google pro users. Um, and it would be $10 for non Google subscribers per month for, for having that. But so yeah, it's another way to, that they can lock in users into their Gemini ecosystem. Right. Um, next up we have Neuralink showing how the operations would happen. Um, the insurgents of the Neuralink by a robot. Let's just play this video and see how beautiful and detailed, um, this thing will work. Um, I personally would prefer a robot putting something in my brain than a human, but I've never really saw how this would happen. Um, any look at you, you're the science corner guy, right? Um, now like Neuralink is also making quite a lot of, uh, progressions. It seems. Yeah. I think I'm reading, uh, online that basically 21 people now in four countries have a Neuralink chip. They're able to type at 40 words per minute, which, um, I don't know if you guys have ever done a typing test, but most people basically can't do even 40 words per minute. So, um, this thing is unlocking the ability to basically interact with a computer today. Um, and it's been reliably tested. I mean, look on, on a small set of people, but nothing wrong has happened so far, uh, at least that has been like broadly publicized. Um, that's an amazing, exciting thing for, I think like the future of computing. And I know we've talked about a lot, like what is the next form of computing look like? Is it, uh, going to be some sort of, you know, Apple, uh, Apple has the vision pro or is it the, the glasses? But to me, this, this type of thing, at least on the input side, I don't know about you guys, but I get so frustrated sometimes with my phone and being like, why am I using my fingers to type? Surely there's a better way to be able to input stuff into the computers. If Neuralink has even figured out the input way to be able to get things into a device from your mind in a reliable and fast way, that's an incredibly exciting thing. Um, and I, I think we like said that the last valuation was something like $43 billion. I would be investing in that, right? I don't know of any other company that's doing that's gotten this far, far along. And, um, this video also just shows how, how, how precisely they're able to insert the electrons kind of like the, the receptors into the brain. Um, yeah, it's, it's, it's incredible. It's, it's, I mean, for now it's only kind of, yeah. Anyway, that's it. Yeah. Incredible. The next thing from its timeline, it's still obviously a medical device. So it's still being used in the medical treatments that are being tested, but these are just the stepping stones, right? So first this technology needs to be proved to be, uh, effective at, uh, at, at, at curing certain disabilities, uh, but conceptually this can then be expanded into the consumer market. That is a bit science fiction. Um, but from the timelines of it, I mean, it seems on track and I mean, it looks like in our lifetime, we might have this as an option to, to, to access our devices. Um, yeah. Then the question becomes like, would you, would you have your kids plant this in? Um, maybe this is a question for Alan. I think you're closest to having kids, uh, from, from us four. Um, would you like, because if you compare it, like the kids that have this implemented and they figured out the input thing, how to input something from your brain into a computer, the kids that have this will just be so far ahead of kids that don't or like people as well. Right? Like if you have to compete with someone on an interview for a job and you have a neural link, I would a hundred percent hire the neural link kid, you know, is that fair competition? Yeah. I should say me and Maria are not planning on kids, but on the, on the, um, all this it's, it's to me, it's fascinating this because it's like already our phones are like this extension of us. We're like almost these cyborgs. Now technology is going into our brains to what extent could you have your own world model in your own brain at some point? If you know, um, Jevon's paradox applies to, to the size of chips, you know, and you know, we, we get, we get embedded into our brain and suddenly you have, you, you implant the ability to have 200 IQ into your brain. This is all extreme sci-fi stuff that we just never would have thought of when we were kids. Right. And now it's just happening. So yeah, I'm with you. I would hire the, the cyborg to do my knowledge work or to be my employees. And no doubt. Exactly. And, uh, we, we stay in the sci-fi because SpaceX again, Elon, they started, I think, what is this? The, they started testing the full trust 33 engine static fire with super heavy V3. And I think this is what's going to power the starship. So this is actually what's going to send those million people to Mars so that Elon gets his paycheck. Um, it's, it's an impressive video. If you look at it, like the static, how this is just a piece of art. Look at that rocket firing. This looks AI generated, honestly, like this doesn't look real. Like what the fuck is happening here? No way. Right. Why is we watching the rocket launch? No, it's not a rocket launch. This is them testing the thrusters that are going to power the big, the biggest rocket ever built. And that's, that's gonna bring us to Mars in a few years from now. Um, there's another viewpoint from, from the top. Look at this. Boom. Trusts out. It's crazy to me that this company that we're watching is also striking a deal with Anthropic to sell GPUs, right? Like it's wild to me that this is the same company that we can like call it the same organization and that they have the same overarching goal. Like on the surface, that just makes no sense. When you break it down into these little pieces and components, it makes some amount of sense, but, um, it's just such a grand vision. It almost reminds me of kind of Kalanick, right? We've talked about the atoms and like this kind of transport and then the data centers aspect. Really. I think Elon is building with SpaceX, the ultimate atoms company, um, from people to transport us from this world, the bits to atoms, right? Is this proof that if you can think it, you can do it. And Elon is just the living example of thinking or believing in something as possible and then actually designing everything around it to make it happen. Because like, that's sort of what it looks like. The interview that I sent you guys yesterday, I think it was a 2007 interview from, from Elon where he is talking about bringing people to Mars, where he is talking about, um, his future vision 19 years ago, almost 20 years ago now. Right. And now you see it playing out. So like, he's been planning all of this. Um, I think since his early days, to be honest, the, the, the interviewer asked him, like, what's your vision of the future or do your companies? Like you already disrupted sort of the, the internet and internet payments. What's next. And he's like interplanetary and energy and boom, here he is. Right. Um, he did it. If you can think it, you can do it. And the most entertaining scenario is probably what's, what's gonna, what's gonna happen. Last story of the day guys. And, um, we have a very, very long here. I can understand that you're tired, but we have to address this. Half stock, half cash. Have you seen the meme? Have you heard the meme? Half stock, half cash. Let me just find the exact video so we can just play it first. And then let's discuss half, half, half sock, half cash. Did you see my meme by the way? So what's that? They made the meme? Half cash, half sock. Yeah. Then I changed it to half cash, half pork. Anyway. And then, and then the cash that gets you to 20, you have this letter from TD. That's another 20. Uh, we're now at 40, uh, but we're still off, uh, by call it, uh, 16. And, and the 20, as far as I understand, while it's considered a highly confident letter, meaning TD saying they're highly confident, uh, that they would provide the financing. It's not locked financing. Yeah. We'll see what happens. What an answer. I hear you. I understand that. I'm, I'm just trying to understand where the, the rest of the money would come from cash, half stock. I, I, I, I hear you. I'm just saying that that math doesn't get you to the, to the price that you're offering. Sorry. That's a pretty straightforward question. I don't get it. Like, where's the rest of the money coming from? Andrew laid it out pretty clearly. I don't understand your question. We're offering half cash, half stock, and we have the ability to issue stock in order to get the deal done. But the full details of the offer are on our website. So this video has been going viral. A little bit of context is this guy is the GameStop CME. Um, yeah, the GameStop GME CEO, Ryan Cohen, and he made an offer basically to buy eBay. The funny side context of the context is that eBay has a bigger market cap or a higher value than GME actually has. And then he is saying that he's going to offer half cash, half stock. And it, the math doesn't really make sense. Now, this has been going super viral and the whole timeline, at least my timeline was full of this. And I found it very fascinating. And this guy managed to just stay top of mind for a few days. Um, looking at he, after this interview, he'd done a bunch of other interviews and then he started selling stuff on T on eBay to pay for eBay. So, so like he's just living the meme right now, right? The video was a meme. Then he is tweeting the meme. He's generating millions and millions of views on his tweets. Um, if you look at it, Ryan Cohen is like, you can see it from a perspective. He's a marketing genius currently trending with over 100 million views. GameStop's own post has generated nearly 60 million views while the new eBay proposal is estimated to have driven over 40 million additional views across social media. Um, it's going viral everywhere. It's trending on Google. It's trending. Everything is trending guys. What are your opinions here before I share mine? I just think the, almost my favorite takeaway from that is when a journalist asks you to report something that they should be reporting, just tell them to read your website and don't be there. And don't be there parrot. Essentially. I thought that his, uh, way of handling the interview was very much controlled and, uh, tactical and, um, very often journalists try to manipulate you to get you to say their sound bites that they want you to say. And he was very effective at making them say what he wants them to say. So that's my takeaway. I don't know. I don't know about how much of a valid offer or or how much substance there is to what's going on. But yeah, the, the, the viral marketing side was, was clearly because he knew exactly what he was doing and calculated every sentence. I, I, yeah, the, the more I look into this, first of all, Rick, if you notice what his handle is, Ryan 50, 50, why do we think it's 50, 50? Half cash. Yeah, exactly. Yeah, yeah, yeah. Which I, which I just picked up on, which is hilarious. So like everything was crafted. Like the more I look into this, this was like a genius, uh, well thought out thing by someone who really understands, um, you know, like how organic marketing, right? Yeah. Organic marketing, the attention economy, all of that stuff is right. Like he's fully leaned into this. And I mean, I went down like his Wikipedia rabbit hole, right? This guy was one of the co-founders of chewy.com. Um, chewy is like a, the Amazon for pets basically. And then I think at 2018, I don't know if something happened, but he left the company there. I'm not sure if he was forced out or he left, but this guy like understands how to run a business, but then come 2021, 2022, when we were in the Zerp era and everyone was trying to buy every type of collectible, he aped into eBay, right? And he put in like, uh, he became one of the biggest shareholders. I want to say he owned five to 10% of the company. Like this was while wall street bets was, you know, really hyping up that whole company GameStop. And then the more I kind of listen and like, think about this like thesis. Well, eBay is where Pokemon cards are being sold. Trading cards are being sold all of these collectibles, right? So GameStop, while they sell, you know, video games and certain, let's say physical assets that are potentially collectibles, you know, like a very old PlayStation could be worth something. Really. I think Ryan Cohen understands this meme culture extremely well. And he sees that eBay is the middleman, the intermediary for this whole, uh, kind of meme economy. Um, he doesn't see it going away in the future. A few things that I did notice is like, he uses the word I a lot in interviews and he's got kind of this Luca, you mentioned in our group chat, this Machiavellian complex to him a little bit. Um, but I gotta like tip my hat to him. Cause like, uh, GameStop is probably only going to benefit from this. Um, maybe they'll get into some legal troubles, but I think the, the attention that, you know, GameStop got from this. Yeah, go ahead. I think, I think like I've been building this case in my head for the past few days and I think I see it here. Um, there is a lot of speculation on the timeline about numbers as well. GME, roaring kitty, not going to go into that yet, but if we look at it, um, so he has been playing on eBay. He's been selling his stuff. He's been generating millions and millions and millions of views, but his whole point is that the current CapEx and even just the marketing spend of eBay, they spent $3.2 billion on marketing and their platform only grew by a million users of this marketing spend. Meaning eBay spent $3,200 per new user acquired. Doesn't even matter if the user actually bought something on eBay. User acquisition costs 3.2 billion, 3.2 million, 3.2, $3,200 per user. Like run a paper, pay-per-click campaign on Facebook, right? You probably do better. So his whole standpoint is I can run this business way more efficient. He probably generated more views with his tweets that he did. Then the whole marketing campaign that eBay did for 3.2 billion actually generated more people went to sign up and to buy and to bid on his stuff. Um, his, like his, he, he posts something on eBay and it just gets bid, um, which is insane. But then what you were saying there as well, Ben, um, I think he sees the value in bringing digital and physical together where eBay is the auction and GameStop has the physical locations. Now see this scenario. I'm, I'm a collector. I have stuff that I want to sell. I don't really know how to get started. I go to a, to a GameStop place and I say, Hey, I want to sell this Pokemon card. I go there and I expect to get $50 for it. Right? I think, Oh, it's a little bit valuable. Now, if GameStop and eBay are one, they can literally throw it on auction straight away. And this person could walk out to the thousand bucks because online bids just flow in on that card. Right? So you have this mechanism, bring online and offline together, which is like, I think that's a genius thing and you don't need much marketing for that. You just need to crypto space to pick up on that. Basically right now, like the whole, everyone who was collecting NFTs back in the days is now on Pokemon cards. Right? And this, you can, you can buy it, you can gamble on them online. You can then get them sent to your, to your place, but all of that. Anyway, that's one side. So that's one side of the story, bridging digital and physical. And the second thing is apparently they've been secretly stacking up huge derivative positions in eBay way before they announced it, racking up hundreds of millions behind the scenes. Now, they only bought 25,000 shares, but they bought close to 25 million shares as derivative contracts. So just alone by announcing this news, they're already up like a hundred billion, I think, into a hundred million in, in value and unrealized profits, just by this little game that he's playing here. Right? So he's winning anyway. It doesn't really matter what the outcome is. He is winning. Um, and then I think Ben pointed out as well, that he has a certain pay package to like, he's not really getting paid anything except for if he hits certain targets and the targets are that, um, it's certain revenue targets. So he doesn't really matter that, um, stocks get diluted, um, which may be not good for the shareholders, but it's like in general, he can only win in this whole thing and he's kind of playing it out. One of the last things, um, is, um, yeah, this is, this is another explainer anyway, that people are, where is it? Oh, it's here. New screen. People are speculating on the date of, oh, on the date, May 12th and May 13th could be a significant date for GameStop. This marks two years since Roaring Kitty's return and Ryan Cohen's eBay listings that he's been publishing come to an end at the same day. If you remember this meme two years ago, Roaring Kitty, the guy who did the whole, who, um, started the whole GameStop saga on Reddit, this was in 2021, no 2020 even, December, 2020. He returned in 2024 and GameStop ripped, right? On the 12th and the 13th of May, this has been two years ago. And this is the exact same date that Ryan's listings expire on eBay. Now, what does Ryan know that we don't? What does Ryan know that the world doesn't know? He has something behind his sleeve. He knows something is going to happen and he is expecting the stock price to hit the price he needs to be able to pay half stock, half cash. And that's what I'm expecting. So I'm very excited for the date 12th, 13th, May, maybe also a bit scared because I don't know what's coming, but it's just like, there's so many data points. Oh, and the last data point is something with Trump. Somehow, somehow Trump is involved here. Um, but a perfect circle. Someone figured this out. Ryan Cohen thinks two years ahead. Ryan posts Trump 665 times. 65 days from seven, from the 17th of July lands on May 13. So another confirmation that May 13, something might happen. This was his tweet. People going to find out, I don't know. But um, yeah, it's just too many things. This is all true. Ryan is the weirdest person I've met in my life. I've heard of in my life. This is way too, too, too, too Illuminati, but like, um, exactly. Yeah. But that's why I like, I do agree with your point that it was like a no lose scenario for him to do this because he, he stacked up on the equity, ripped the, the news and the viral trend out there. If he succeeds, GameStop acquires eBay. If he doesn't, he pumps stock that he owns. Yeah. Yeah. And if he gets the, and if he buys eBay in the end and he saves $3 billion in marketing spend on a yearly basis, that's $3 billion that could be put into whatever cash reserves, um, that could yield, that you could yield instead of being spent. Um, and then also cutting out 90% of the staff that's not useful in this whole business that he, he is literally saying I can run this from my bedroom. Um, well, yeah, Ryan, um, I want to see what's going to happen on May 12th, May 13th. I'm, I'm excited. Maybe we should dabble into some GameStop stock, uh, in the meantime, uh, but yeah, not financial advice. I believe in the idea of merging physical and digital. I believe in the collectors. I was part of the NFT craze. I unfortunately didn't, didn't really jump out at the right time, but Hey, it made me see that there is a lot of, um, value in scarce assets. And yeah, maybe one more thing is that indeed, like the, the kids from now, they much rather buy vintage clothes, for example, than, than going to Zara. And that's another thing, right? Like you're buying it on eBay. It gets sent to, to GameStop locations. You can go and test it out. Like you don't really have to trust a rando to send it to your door and then maybe it doesn't fit or it's not like it's advertised. So like, that's just a little addition, but Hey, I can continue talking about this. We've, we are two hours in guys. This was the God mode pod episode 12. Hopefully this one hits a thousand views on YouTube. That would be, um, in line with our growth trajectory because I am the greatest moderator and I have the best co-host. Thank you friend of the pod, Alan for joining us and staying for so long. I think it's already late in Philippines. So it's bad time for you. Um, and yeah, we see you guys next week. Bye, bye, bye. Bye. Ciao. Cheers. All right. Coding an app is the new starting a podcast. We are three dudes. We just started the podcast. It's incredibly reckless in this world to not learn AI. Everything is code. I recently realized everything is code. Luca, what are you predicting? Underwater drones run by people. Okay. Wow. Bombshell. Use an AI to write your promise, bro. Come on. I am the greatest moderator in the world and we see you guys next week. Love you besties.