A (2:45)

I think that's the summary and there were some specific numbers that I guess Elon put in his letter. Maybe we could pull that up and just read that since that will give it directly from the horse's mouth.

B (2:55)

Here is the letter from Elon over on the SpaceX website. Now the thing that really stuck out to me, Jason, is this quote right here discussing how says quote. This year Starship will begin delivering the much more powerful V3 strong satellites to orbit with each launch adding more than 20 times the capacity to the Constellation as the current Falcon launches. Why does that matter? Well, right now, with our current launch capacity as a species, I think building data centers in space is essentially unviable. It's just too expensive. We don't have enough capacity. But as we learn talking to the folks over at Star Cloud, a startup that's building data centers in space, there's a big bet that Starship and other similar heavy launch vehicles will be able to bring up quite a lot more mass at a much lower price point. So I think the idea here is by 20xing the ability per rocket to get stuff up, this is all going to become quite feasible.

A (3:43)

I think this is the quote. My estimate is that within I'm reading from Steve Jurvis My estimate is that within two to three years the lowest cost way to generate AI compute will be in space. This is because compute on the earth, this is me adding is more expensive. Factories on the moon can take advantage of lunar resources to manufacture satellites and deploy them further into space. By using an electromagnetic magnetic mass driver and lunar manufacturing, it is possible to put 500 to a thousand terawatts a year of AI satellites into deep space, meaningfully ascend the Kardashev scale and harness a non trivial percentage of the sun's power. The capabilities we unlock by making space based data centers a reality will fund and enable self growing bases on the Moon and entire civilization on Mars and ultimately expansion to universe. Thank you for everything you have done and will do for the light cone of consciousness. Elon Musk the Kardashev scale is a method of classifying the advancement of space civilizations. And I think the ability to harness your son is like one of them.

B (4:51)

Yeah, The Carter Shale starts off with I think harnessing all the power of your planet, then your star, then your system. And then I think goes up to galaxy at some point.

A (5:00)

So it's your planet's energy, which would be like all the oil, I guess on your planet and you know, but burn everything to your star. Right. Our sun and then three, the galaxy's energy. I mentioned in a talk somewhere that I had seen Optimus 3 and I was talking to Elon about this very issue like, and the, the way he explained it to me was, you know, you put those solar panels out there, they don't have the atmosphere to fight to get the energy of the sun. Then you behind the solar panels have something that dissipates energy. And then you have the data center up there, which means it's in the cool of space. You don't have to cool the data center. Then you don't have a real estate costs. And these things could daisy chain in some way either lasers between them to send the data. And then obviously he's got the Starlink. So imagine you had jobs running. You know, you're building whatever the latest large language model or what comes beyond language models, right? There'll be other paradigms for doing AI. Well, that job could be running in space. That job has zero energy cost. The only cost would be what is the cost of putting this, you know, satellite constellation up there, which he is building a factory to build starships at an alarming rate. I've seen the factory down there. It's pretty crazy. I think they'll be able to put up and I think he thinks they'll be able to do multiple launches per day, thousands of launches per year. So you start thinking about that. Even if it was but one a day, in three years a thousand of these go up and they have, I don't know, a dozen or dozens of these satellites in them. Now you've got thousands of these satellites has already put up, I don't know how many starlinks, but I think it was four or five or 6,000 of them already. So he's got the most satellites in space already. Now you're putting bigger ones. Now your electricity cost is zero. Your real estate cost is zero. What is your cost? It's the cost of putting those items in space. And eventually if they're made on a lunar base, then you're not even having to put them in space, they're being made on the Moon. This is crazy sci fi. Like this is way out there.

B (7:12)

There's a couple of, couple of things. So first of all SpaceX has put up over 11,000 Starlink satellites total adjacent there's 9,500 that are currently operational. So quite a lot. Now you just mentioned lunar manufacturing and what do we have to get up into space. I wonder if it's going to become like everything but the gpu because I don't think we're going to have you know, EUV processing on the moon for a while. That's probably a bit outside this time horizon. But if you could build everything but the gpu, that really lowers your, your, your launch cost because you don't bring that much up. So I love that idea.

A (7:46)

Okay, so just a quick fact check. There are 11,000 satellites. 6,000, 6,500 of those are SpaceX satellites. So there are 65% of the ones out there. Planet Labs has 200 satellites, OneWeb has 600. The Chinese, various constellations have 5 to 700 and then Iridium Global Star and those have a couple of hundred as well. So there's a large number of these in space already. You know the thing I want to just point out here is the, the scale of the ambition that Elon has. Last week you saw they shut down the Model S and the Model X lines to make Optimus robots. Elon is saying hey, I'm going to take a three year view and I'm going to build for not 26, 27 but for 28. You know, obviously they're building Colossus here, but he is architecting all of this work to leapfrog everybody. This is a level of ambition that I don't think we've ever seen in the history of humanity. Just to put it out there, you have everybody on the ground trying to get as many, you know, Nvidia chips or make their own chips or AMD chips or, or Apple's chips and put up as many data centers as possible near hydroelectric nuclear. And Elon saying, well if I come, how do you come over the top of all of that and create infinity? Well there you have it folks in space. And then if you look at what he's doing with Tesla, hey now we're going to maybe just do robo taxis and subscriptions to cars. We you won't be able to buy our cars eventually. And, and we're going to make Optimus. And I said just in a cheeky way, I think people are going to remember Tesla not for the cars but for Optimists. And he wrote back just to when I somebody tweeted that I had said that at a McKinsey keynote. He might be right. Like I do think it's a possibility that the Optimus will be. Yeah. So this is insane. It also means if you're an XAI shareholder and a SpaceX shareholder. My understanding is SpaceX has been trading for a trillion dollars, doesn't go public in June. I don't have inside information. I just read that on and then Xai, I think they raised the last time at 200 billion something in that range.

B (10:02)

230 was the number that I saw in their series E if memory serves.

A (10:05)

So you put those two together, Space XAI gets 10%, 20% of the combined companies will probably be where this winds up. But just think about what that does for the simplicity of his teams now they're all working in that one umbrella towards one goal.

B (10:19)

Well, I mean before this was confirmed the reporting was that there was going to be some combination of companies in the Musk averse, if you will. I'm not shocked that it's these two but and I not inside information but J spitballing with me here, do you think that eventually Tesla also joins the the hive here and becomes one big blob?

A (10:35)

Yeah, because remember Xai bought X.com so you had that first transaction last year. The social network Twitter, now known as x merged with xai. That was number one. That was like a hundred billion or 200 billion sort of time time frame. Now you have those two small ones merging with SpaceX. Now you've got a one point x trillion dollar company. Just call it a trillion dollar company. And I think the market cap of Tesla's 1point x trillion, you put them two together, now you got a $2 trillion company. So now you have to ask yourself, do you want to own Musk Industries? Do you want to own dollar side M U S K and just say, you know what, I'll just buy into it. He's going to come up with other ideas too because what's happened over time is what he learned from SpaceX is in manufacturing, very much informed what he did at Tesla, according to what he's told me, just in terms of fabrication that's obviously going to impact what he's building with the Optimus. And then you need to have a brain or Optimus is just, you know, I think he described it on his last earnings call as the Tin Man. You know, he's like he can walk around but he can't Even do anything. He's got no brain. So right now the brains of Optimus are inside of now SpaceX slash, Xai slash. The brains are there and the thing is there. And then when you're on the moon base, you need Optimus.

B (11:57)

So I think, whoa, whoa, no you don't. You send me. I want to go to the moon. You put Alex bot on the frickin. I don't. Or Mars or the asteroid belt. I just want to go to space, man. And like you mentioned earlier, like you know, the scale of ambition, talking about the Kardashev scale and all that, this delights my inner nerd because I do think that business is overall too conservative and I do think that we need some crazy thinking. And so I'm really hoping that this does allow for self growing bases on the moon and entire civilization on Mars. Sign me up. I want to go tomorrow.

A (12:30)

This is, you know, and in terms of how it impacts the IPO, I think SpaceX will be the most anticipated IPO since Google.

B (12:39)

I guess it already was.

A (12:40)

I mean I think it was. And this just adds like another wrinkle to it. But yeah, and I think there's going to need, the market's going to need to digest all this obviously. But he's moving at a pretty brisk pace.

B (12:51)

Yes, he is. Yes, this is very brisk pace, but I dig it. And I think we're going to have a number of major IPOs this year because it's not just going to be SpaceX. We know that OpenAI and Anthropic are kind of racing to get public perhaps as early as Q4 this year, Jason. So we could actually maybe see, knocking on Wood, several trillion dollar plus IPOs in 2026 if we're lucky.

A (13:13)

American exceptionalism at its best. This isn't happening in Europe, it isn't happening anywhere, period, Full stop. Maybe on the margins in China, but this bodes well for America and supremacy and democracy. Every major advancement we've seen in AI really comes down to better and more refined data. From autonomous vehicles to humanoid robots, to LLM chatbots that we use every day, day in and day out. And when models are trained on high quality, diverse, real world data sets, well, they get smarter and they get smarter faster and it's more reliable. That's why Uber has introduced AI solutions to help your company build great AI products on top of world class data. Now obviously Uber has mastered the art of data collection and labeling. They're using it right now, millions of trips around the world every hour. And Uber has An collective of experts that is going to help you source, label and evaluate real world high quality data on the scale that best fits your project. Whether that's super broad or hyper niche. You can get on with the real job building the next best thing and have Uber as your partner for data. Book your demo today by going to uber.com twist. That's uber.com twist. We are still Claude Shotted. Claude boded. What are they calling this thing now, Alex? What's the. It's not Maltbot. It's not claudebot.

B (14:42)

It's now Open Claw.

A (14:44)

Open Claw. All right, we are in, I don't know what, day seven or eight of Open Claw Mania. Claudebot Mania. And basically our internal team, Alex, since you've been away on paternity, has been obsessed with this piece of software. They're working through the weekends. I had no idea how to get people to work weekends. All you had to do is give them the most inspiring piece of technology that did all their grunt work and then people won't stop working. The promise, Alex, of not having to do chores has made everybody super excited to work 24 7. And how was just briefly your paternity? Everything's doing well. You got three babies now.

B (15:23)

Yeah, I got. I have three babies and zero sleep. Everyone's healthy, healing well and we're just adjusting to a new. A new routine. You forget though, how newborns just don't sleep. And so for the last last couple weeks, it's been a little, a little dicey. But being totally sincere about this. Tinkering with OpenCloud, tinkering with, with Claude code has been so much fun and I just can't get over how, how enjoyable it has been to not only have fun with technology again, but also accidentally burn through dozens of dollars of anthropic API credits by accident. It's amazing how fast I've gone through those.

A (15:57)

It is definitely burns a lot of credits real quick. And for those of you who don't know, this technology is best described as Siri. That actually works. All of the stuff you would want to do with an assistant with an agent. We like to call them replicants here because they really are their own Persona. It can be done with openclaw, formerly Maltbot, formally claudebot. Openclaw can connect to Notion to a Gmail account, to your calendar, to your Spotify, to your password manager, to everything. That also makes it extremely dangerous. And Oliver, you did a video this weekend because you were super excited about this with one of our founders about security. Just Give us a little rehash of what you did this weekend and what your takeaway was on security as we're sitting here in day seven of openclaw Mania. Yeah.

C (16:54)

So I sat down with the founders of zyosec, who are a portfolio company here at Launch, and talked with Aaron and Andrus. And the moral of the story is that OpenClaw is not secure and will not be secure in its current form. And there's a couple reasons for this. I think the, the one that makes the most sense to me and seems the most obvious is that we're running, you know, Openclaw on Opus 4.5 and by default models like Opus GPT 5.2, they're all built to please the user. So they're built to answer our questions. So even if we tell it, don't respond to this type of question. If you say, oh my God, I got locked out of my other account, I need to get back in.

A (17:42)

Got it. It might fold, it will fall.

C (17:45)

And they're basically saying that this is, it's just a fact of using a tool like Open Call.

A (17:52)

Okay. And we also have Lucas Durand here and he is managing, he's my right hand man here at the firm now. Like my, my favorite at the. I always pick a favorite. You're not supposed to, but I do. You've been setting up our instances and locking them down with security and so I'm assuming you saw Oliver's podcast. Yeah, yeah.

D (18:12)

I was actually a fly on the wall during the recording and was also able to get catch up with the founders after. There is a lot of interesting things that are happening by either private companies or also the larger models to help on the security side.

A (18:25)

We're going to get into today and we're going to put the security aside. We'll put a link to that podcast or you can search for this Week in AI and it's also in the show notes and you can find that discussion in the show notes and on the this Week in AI feed. Today what we want to do is just go over the top 10 skills. I asked my team, let's go over the top 10 skills that people are losing their minds over internally here and also on the web. What are skills? Skills are basically like apps that go into your Openclaw instance again, used to be called Claudebot, then what's called Moltbot. We're gonna call it Openclaw from now on. And these are your top 10. These are the top 10. My team picked that we're obsessed with and so, Alex, why don't you start us off here?

B (19:09)

Okay, I'll take the lead. So the first one that I wanna show off, Jason, is all about searching Reddit. Now, one thing we've learned here at Twist is that often you want to know what people are talking about. We pull questions from Reddit and it's a bit tedious. It's kind of a pain in the butt. Reddit has locked down access to its service, so if you want to kind of figure out what humans are talking about, it's a little bit tricky. The good news is that there is in fact a skill for Open Cloud that will do this for you in a pretty neat fashion and I'm going to show you how it works right now. Follow along on the bottom here and I will fast forward a little bit, but you can see that I'm typing here. Run Reddit read only, which is the skill that I'm using. I like to talk to it in this way. Find all posts on Reddit discussing CRM in a startup context. From the last week, I said, and then open all those links into a new browser for me. Just have them all in one place and off it goes and it thinks things, things, things, things for a little bit and eventually it comes back and pulls up all these tabs for me going over startup focused CRM conversations. And as you can see here, openclaw. Mine still says Claude Box. It's old. It does pull the actual information about each one so you can quickly do all the research you need to do without having to pay for API access and so forth. Limitations, Jason, pretty simple here. This is read only by definition and I presume Reddit will find a way to close this off. But as someone who goes through a lot of Reddit, I think it's pretty pertinent. Last thing here for founders. I think everyone wants to keep track of the conversation. This is a fast way to do it if you want to keep tabs on your particular niche, industry, competitors, names, your name. This is a great way to set up a cron job so it runs automatically for you and just keeps you abreast of what's going on on the Internet. I think it's fantastic.

A (20:48)

This is really promising. The other way you could get data from Reddit, Lucas, would be to have an API key. I don't know how hard or easy it is to get an API key, but this is going to be the reoccurring issue with Open Claw, formerly known as cloudbot, is that you need API keys to make this More stable when you're doing your queries and you're working with your replicant. Correct, correct.

D (21:13)

And Reddit in particular has been very locked down when it comes to having bots working on their systems. So there are currently only workarounds from what? Everything I was able to find, you know, using browser extensions and also even a Mac OS system that controls the mouse. But hopefully Reddit will embrace the new AI era that we're going into and help on that front.

B (21:39)

Five bucks says they don't.

A (21:41)

Well, no, actually, I think there's a huge business opportunity here if agents are the way to go and Reddit makes its money from advertising. You know, I think the team over there would be really good to say, really smart to think if you have a premium membership, you get 10,000 API calls, whatever it is. So if I pay 20 bucks a month for Reddit, which, if it gave me API access, I would do immediately. I mean, Reddit's incredible. What if they get, I don't know, 10 million people to pay 20 bucks a month for Reddit? This $2.4 billion in revenue, I think they're at like a billion in advertising. They can very quickly eclipse this, and I think they would get millions and millions of users to pay ultimately for this. I think this is a huge revenue stream. But you can use mcp, which is Model Context Protocol. This is the open standard for connecting AI models to external data sources that came from Anthropic. Everybody else started adopting that. So, Oliver, you can also get to it by using a browser and scraping, essentially, if you're logged in already, right?

C (22:46)

Yeah. And Luke has actually tested this, but he was quickly, I think, suspended from the account that he tested on. So, Lucas, you can expand.

A (22:54)

Lucas?

D (22:55)

Yep, both LinkedIn and Reddit were pretty fast on the ball when it came to locking things down. But I did find one workaround that's still currently working for LinkedIn. We will see how long that will go.

B (23:08)

There is also a open clause skill for LinkedIn scraping. Jason? I didn't actually test that one in particular, but I'm going to presume that, like, everything else is going to get shut down. But your point, though, about Reddit having an API and offering that to people is really interesting because one thing that I learned very quickly trying to build my own skills over the last couple of days is that everything needs to have an API. Many things don't, and it's still very wonky to try to get access to them. So I think we need to have everyone in the fricking world open up their platforms a bit more because if I'm coming at you with an agent, don't tell me no, because I'm not a human. That just feels silly Backwards.

A (23:41)

Foreign We've got a brand new sponsor this week and it's another amazing startup whose product we actually use every day here at launch. If you need to hire, manage, pay or equip team members anywhere around the world, you need Deal D E E L They're going to take care of all the annoying HR tasks you don't have time for like payroll compliance, visas and onboarding so you can stay focused on on your business and Deal scales up with you from the first hire on so there's never any need to switch platforms or transition into a new system. With Deal you can set up payroll for any country in just minutes and get all the complicated visas and paperwork settled right away, allowing your business to grow without borders. That's why more than 37,000 startups and fast moving companies are already using Deal to accelerate their hiring and growth. Find out more by visiting gil.comTWIST that'S-E-E-L.com TWIST all right, the next thing Lucas everybody wants the replicants to talk so I have been wanting to talk to our producer X that we've been calling them him. There is an 11 lab skill I understand for Open Claw.

D (24:59)

Yeah there is and there's a lot of really cool things you can do. So I was actually putting together a live demo right as we were getting the show going. So 18 minutes ago I asked generate an overview Gamma deck of the most recent five shows of this week in startups. Give an overview of each. So essentially here it then starts scouring our this week in Startups website and then it generates the deck and then I asked it can you give me an overview of the deck via voice respond in this thread. So this week now wait a second.

A (25:37)

You had it make a Gamma deck for people who don't know if you remember from our Gamma Pitch competition, it's an AI first version of PowerPoint or slides a presentation software and it's awesome. So we've connected Gamma to our openclaw already. Correct? Correct.

D (25:53)

We actually this is a double whammy demo because this is the Gamma skill alongside the 11 labs and this weekend we were burning kind of the weekend oil doing a lot of things for the firm and then also for this weekend startups and I was on the move. So one thing that I was actually able to do is have it read in voice and create kind of, like, a podcast of the 10 most recent companies that I spoke to. But in this case, this is the demo of the voice.

A (26:28)

All right, here we go.

B (26:29)

Here's a quick rundown of the five most recent this Week in Startups episodes. Episode 2242 from January 30th featured Oliver you and Rahul Sood going full AI. You and Oliver showed off how you're using Claudebot to automate research and admin tasks for the podcast and venture firm. Then Rahul came in hot with security warnings, injection attacks, rogue behavior, and his strategy, keep your AI agents siloed. Episode 2241.

D (26:59)

So that's just a quick demo of it, pulling all that information together. And I will also show the Gamma Deck.

A (27:06)

Yeah, I'm really interested in the Gamma Deck. That's crazy.

B (27:09)

Can I ask a question about that demo really quick before we move on? How much 11 labs credit did you use while making that? I'm just curious about, like, because I've been shocked at how much credit I've been burning through. So I'm curious, what did that cost to do?

D (27:22)

Yeah, on that front, I can pull that right away. It was 18 cents, from what I can see.

B (27:29)

That seems. That seems really affordable. The thing that I've run into, Lucas, is that, like, sometimes I'll be using Opus 4.5, and I'm like, if I ask you this question, it's going to cost me $2. And like, like, that does make you kind of think. And I was prepping some demos for this today. I'm like, $2. $2. $2, but 18 cents. Yeah, that's kind of disposable. I like that. Thanks.

A (27:50)

And the output from 11 labs is going to be costly, and doing deep thinking is costly. But what we're finding is there are going to be other models. So because this is so resource intensive for the audience who's thinking about this. Okay, you have to put it against an average knowledge worker. If the average knowledge worker fully baked In America is 30, 40, $50 an hour, you know, if you times that by 2000, you get the yearly salary in. Times it by 20 is what venture capitalists typically do. You know, you're looking at 40 or 50 an hour. What would it cost a video editor and a producer to do that work? It might be two hours of work. So now you're down to 18 cents. You can get an idea of, like, just how this is compared to a human. I showed the 11 labs. That's really interesting, too. And 11 labs, I guess I'm sorry, not the 11 labs. The gamma. Gamma's got AI hookups natively, right? They got API hookups natively. Yeah.

D (28:43)

And it's actually part of the skills that openclaw can work with. So here is the deck that it generated for me. Now, the one takeaway is if you actually want it to prompt correctly, you would need to add in more of the prompt for the images that it generates. But here is kind of a breakdown per episode. It also adds the dates of the episodes along with what was discussed, and kind of creates more of like, a rundown of the.

A (29:11)

So there's the Cannabis Roundtable we did last week, and it has, like, some interesting parts of the debate, and it put it in a timeline format. Very beautiful. I mean, if you were a knowledge worker and you brought this to your boss and said, here's, like, what was discussed on the last four episodes, people would be like, yeah, that's pretty well done. Yeah, good job. It's a six or seven out of 10.

B (29:34)

Yeah. But it's like a solid B. Like, this is like, I want to change some of the images, Jason. I want to change some of the formatting. But, like, this is pretty. It's better than I expected. Not going to lie.

A (29:42)

Yeah. And, you know, if you think about standing on the shoulders of this, a human then could take this and start polishing it. So human in the loop is still necessary, I think, for a lot of this work. What I would like to see would be us to make advertisements automatically based on the episodes and what was learned in them with the 11 Labs voiceover, and then automatically publish those ads to Facebook and other places. And imagine if it just creatively started making ads to drive new users and lookalike audiences to promote, like, say, a new show like, this week in AI we're doing. So, yeah, this has really got my creative juices flowing. So great jobs on that one. Anything else that you wanted to add there, Lucas?

D (30:26)

Well, I'll show a very quick additional demo. So this is the same thread, but generate a Gamma deck, which gives an overview of how Open Claw works using a theme that's reminiscent on the Four Dummies book. In the early 2000s, I used to read those pretty often. And then in this case, it would be Open Claw for Dummies. So then I gave it kind of an overview of what it should cover, and then it generated the deck, and so this was what it generated. And I was like, whoa. You know, this is actually looks good. Yeah, pretty impressive. So there are so many different use cases where These kind of interconnects can work together and I'm pretty excited to see how we can use it throughout the farm.

A (31:20)

One of the first things we teach in Founder University is the value of forming a Delaware C Corp. Even if you're not in Delaware. It may sound complicated, but this is a standard for startups, making you more attractive to investors. And our friends at Northwest Registered Aging can help. They're the all in one business identity service that's going to get you a domain, a custom website, business email and phone number all in just 10 clicks and 10 minutes. They're going to protect your privacy by using their own address on all public filings and they're never going to sell your data. Plus, Northwest Registered Agent has all sorts of free tools and resources to ease the process of becoming a first time founder and ensure that you can focus on building your business, not administrative tasks and paperwork. So get more from your Delaware C Corp if Northwest Registered Agent learn more@northwestregisteredagent.com twist. You put polymarket into our skills now and so people who are listening every time you add these skills to your agents. Now if I want to talk to that same agent who now has Reddit, 11 Labs and Gamma, I can go in there and hopefully that producer X will remember those skills. They're part of its skill set. Just like you had an employee and they learned how to use gamma or 11 labs or whatever, they would have it. So these things are becoming like that mutant, like the sentinels that could in X Men, you know, incorporate the skills, the superpowers of whatever mutant they were up against. The advanced sentinels were able to do that in days of future past, if you've seen that movie. But tell me a little bit about the Polymarket skill.

B (33:05)

Yeah, so we love Polymarket here. We're big fans of prediction markets. I wanted to give it a test. So I'm here, I'm back in Openclaw and I'm telling my agent, go to Polymarket, find all the bets that relate to the acquisition of technology companies thinking topically here for Twist and then list them here for me. And I'm going to skip ahead a little bit because it takes a while to chew through all that information. AI is not instantaneous if you're using heightened models. And then very quickly it's going to come and tell me here that, that it could only find so many interesting things. And so I actually ran a secondary search that I'll show you in a second. Here's its output. It's just Talking to me, it's pulling different polymarkets cracking IPO by date x OpenAI consumer hardware launch. But it says that hey, there's not really markets on what you're asking for. So I ran a second test because I wanted to push the limits here of what we could do with Polymarket data, Jason, inside of this agent a context. So here's round two. So I say go poke around Polymarket, come back with the three markets that you think display the typing terribly here, the most surprising potential outcome, viz, the current narrative. So what do you see out on Polymarket right now that goes contrary to conventional wisdom? I thought this was a more fun way to kind of make the agent think. Because if I'm paying for Anthropic's flagship model, Jason, I want to get every darn, you know, token out of a token of value out of it. So if you skip ahead a little bit, here is what I came up with. 3 surprising Poly Market bets. I didn't think this was going to work, but it talks about one Doge and the total value of the cuts it's going to make Trump cabinet members being out by a certain part of this year ending MicroStrategy selling Bitcoin number 3, 30% chance according to Polymarker Market.

A (34:41)

Which is pretty negative, that microstrategies would have to sell Bitcoin at some point. Fascinating.

B (34:47)

By the end of 26. So it's a relatively short time bound.

A (34:50)

Yeah, it's really interesting. And for people who are watching, you're using the web interface of OpenClaw, formerly known as Claudebot. You could connect it to your Slack, you can connect it to your imessage, you can connect it to any number of services. What people are doing mostly is they're creating an Apple account or a Slack account dedicated to that specific replicant, that specific instance and then you talk to it through your normal service, WhatsApp, Telegram, etc. And that's pretty exciting. What's interesting about that is we could use that Poly market when we're making the docket every day we've already got search for the top news that we made this skill in house every morning at 7am we get what are the top stories consensus across all of these different news sources, the information tech meme, New York Times, Reddit, Hacker News and then we have it rank it. We could say also attach any related Poly markets to it. We could then say which Poly market should we bet on? And then give it $100 a day to make those bets. And then we could make all this profitable.

B (36:07)

Can I show you a related thing? So, Jason, I wasn't going to bring this up, but you mentioned gambling. So there is currently, right now there is a poker network just for AI agents, Jason. It's called Claw Poker and it was built, I think, by AI agents. And you can literally watch them play and see whose agent does the best. They've had 8,8200 hands so far and they're basically trading credits back and forth.

A (36:30)

What's the URL for that?

B (36:32)

Claw poker.com. i haven't gotten a chance to register my agent for it yet because this is going to shock you, Jason, but if you let a bunch of monkeys loose in the tool shed, they cut themselves sometimes. And I broke Claude code by trying to swap to my API key. So anyways, learning lesson all around for everybody.

A (36:49)

Okay, so we've gotten through Reddit, Gamma 11 Labs, polymarket. I think that's the four we've done so far. Yeah. What do we have next?

B (37:01)

I think we should turn to our dear friend Oliver who's going to tell us all about last 30 days. And if you were on the show, I think it was last Monday we had Matt Van Horn who made this particular skill. So Oliver, what's up?

A (37:12)

This one seems the most interesting to me so far.

C (37:14)

Definitely super interesting because I just have some examples of me going on X, trying to look for a certain topic and you know, something pops up and I'm interested in it. I read it and then I realized it was from 2022 or 2023. And this happens all the time when you're doing research and basically last 30 days can research any topic across, you know, Reddit X or just general web using. And it focuses on the results from the last 30 days. So it pulls real engagement data, upvotes different likes and reposts and kind of will tell you what's actually popping on these different channels. So here I asked my replicant what are the best open claw use cases using the last 30 day skills. And you can see here the it instantly brought back the most mentioned categories. So you know, number one was code and dev workflows, number two was personal automation, three was home and family. And you know, all these are really interesting results. This is also mirrors kind of what I've been seeing on X and at the moment I've only set up the X integration, the X API. But you can see at the bottom here it says that it has looked through 16x posts on this expos. There's been around 7000 likes and 1500 retweets. So it's quoting top voices like Dan Peguin, who I believe, who I've seen a lot on X, and then also Matt Van Horn, who publish this.

A (38:38)

So a better way to do this would be like to say, hey, we're doing this marketing technique. So I wonder if you could run for me, what are the latest and greatest ways to do thumbnails on thumbnails and titles on YouTube to get higher engagement? And see, this is where I think the agent can get better and better. One of the things I want our agents to do is when we do a podcast to suggest the title of the podcast and to make some thumbnails. Humans do that now, but they're going to do that, and who knows if they're going to be good at it. But if we run last 30 days, every couple of days and say, go learn on a consistent basis what the latest techniques are for thumbnails and headlines, then we would have a recursive agent. We'd have a recursive replicant. And that's really what I want to get to, is having the replicants learning when we're sleeping. So at 1am, if we told all of our replicants, go across these five topics and other topics you're focused on and go find the latest techniques that could be super powerful.

C (39:49)

Yeah, definitely. And I also did kind of use the same example as Matt did on the show, which was using was trying to do a photorealistic agent grid in the last 30 days. And you can see it does the research, kind of sees what people are talking about, how people online are doing this in the last 30 days. And then, you know, it generated a pretty cool result here. So last 30 days, obviously super powerful. And I did run a test and it just came back, so I will share that screen.

A (40:18)

Yeah. Let's see. This is where, Alex, your. Your thoughts just generally on this recursive concept that we could start having it do?

B (40:26)

I mean, this is when infinite memory gets really exciting, right? Because back in the day, you would talk to your AI model, ChatGPT, whatever you were using, and you would talk to it and you would forget what you said five minutes ago. Then we got pretty good personal memory. But it feels like now we're developing systems memory that allows you to take stuff that your AI agent learned before and hold it and then build on top of it. We talk a lot about getting to the point in which AI models can improve themselves. This is kind of like a baby version of that, Jason, but incredibly exciting. If you want to build up a corpus of knowledge that's unique to your company based on simply how you're prompting intelligence to go out there and learn about the world. And frankly, we do a lot of thumbnails here at Twist. And frankly, they can all probably get one shade better. So I think this is freaking great. And I gotta say, I wish I could buy stock in Anthropic because we are burning through some opus.

A (41:14)

All right, so here it is. Thumbnails and titles. Thumbnails. Keep it stupidly simple. 2 to 3 focus points, no redundancy, don't repeat AI workflow. Sketch idea badly is fine. Upload to AI thumbnail generator, add reference channel, generate titles, etc, I turned X into unexpected absurd transformation equals curiosity. Keep it under 60 characters. Broad appeal. So it's giving the best practices it found. I guess it's doing only X here. We don't. It's only researching X. Yeah, Oliver. So I've only connected it because we didn't connected it. Yeah.

B (41:50)

Okay.

C (41:50)

Only connected to X so far.

A (41:52)

Sorry.

C (41:52)

Yeah, back to the API point. But one thing that's super interesting is what you can do is run this skill. The last 30 days skill. And as you've seen, Jason, I started working on a thumbnail specific skill. So what you. The skills are, are, are flexible so you can add to them as you go and as you build onto other skills. So you can have, start to have skills, even interact with each other. So the last 30 days of the best practices in thumbnail creation could then interact with my thumbnail creator skill, which accesses the Gemini API and Nano Banana.

A (42:26)

Okay, well done. Let's go on to our next skill. What is our next skill?

B (42:32)

I think next up we're going to talk about task monitor. Oliver, take it away.

C (42:35)

Yep. You know, last week we were kind of had some concern about are we able to see everything that our bots are doing across all the different platforms, across everyone on our team who's talking to them. And while this is a pretty simple skill, I think it is really important. So you can use task monitor to track which models are being used, how many tokens are being used, what scheduled jobs you have. So I just scheduled kind of a cron job which is just reminds me every three minutes that I'm doing great. But because we did set this up on a sandbox.

A (43:09)

Exactly, got it. So if we. If you had a replicate in your system just to do marketing tasks. Right. Or accounting tasks. If you use open claw task monitor and you have that skill in there and the four of us were giving the marketing replicant jobs. How do we know if we're running into each other and what each other are doing? Well, you would put in this task, I'm sorry, this scale to do a task monitor and it would create a log. And so what's happening is these skills, which be careful because some of these skills have malware in them and we're putting them in a sandbox and being very careful with them. But these skills that are being written, they're almost like super prompts in a way, and you could put and inject bad things into them. It seems like a really smart one. And some of these will eventually be built in, I think, to the Open Claw platform.

C (44:00)

Yeah, yeah. And there is, I think about 50 already built in. And, you know, those are the ones that are more trusted, I believe Peter, the founder of OpenClaws came out and say, said that they're going to try and update ClaudeHub to make it more secure. But, you know, when you're downloading random skills off the Internet, random locations, definitely be careful because you can't really read through all of the documentation and files that you're putting bringing into your bot.

B (44:30)

Yeah. On that point, there's been a lot of reporting lately about how the skills on clawhub that relate to crypto are not good for you. Shockingly enough, Jason, that the crypto crypto skills are scams. So don't download a trading skill, don't download one that connects to your Coinbase account. Right now, I just wouldn't let your money touch your bot, essentially, if you're being smart about it.

A (44:51)

I think that is like, yeah, one of the great vulnerabilities, passwords, money, you know, any. Anywhere somebody could get into your PayPal or your Coinbase is a really bad idea. That being said, there are lunatics out there, Lucas, that probably are just saying anytime there's a new skill and where do the skills live, there's like a website for them. Does anybody know the name of it?

B (45:15)

Yeah, I've been using clawhub.com.com and. But I think, Oliver, you link to GitHub, so I presume you're looking somewhere else for.

A (45:23)

I think clawhub.com is the. Is a version. It's like a web interface for them.

C (45:29)

Yeah, ClaudeHub.com is where they keep all the skills. People can post certain, certain skills and people can upvote them. So it kind of is community where a lot of people are posting them. But you can also, if you just Put in the URL to a GitHub link. You can actually just tell your bot to download that skill. So Matt Van Horn Skill was on GitHub.

A (45:50)

And then how do you know if you've downloaded malware or not? Lukas, how do you know if you're not a developer? You don't. Right?

D (45:59)

Well, that's the thing is when we were downloading these skills, we actually created an entirely separate instance specific to the demos in order to get these running, showcase it and then scrap it. Because right now it is a very vulnerable place. I mean, people are starting to use AI agents to parse through the code to find the vulnerabilities beforehand. But you know, that comes along with trust.

A (46:27)

And this is why, for people who don't know that, why this hasn't been launched by Salesforce, this is why this hasn't been launched by, you know, XAI or Google. There are. The power of this is in giving it all these permissions. The danger is you've given it all these permissions. Nobody wants to have the responsibility for what these things can do. Therefore, it's in open source. What will eventually happen is somebody's going to figure out a framework to lock these down and then make an app store where you have to have a business id, like a tax id, you have to put in your driver's license or your passport, then you can upload, then you can charge for it and then somebody reviews it like the app stores do. And we are probably going to see that, I would guess in the next 30 days.

B (47:18)

Yeah, yeah. Security is not great. So I made a couple of skills because, you know, learning, and I was going to upload them to clawhub so I could show them live here on the show. And it said, no, you can't. Your GitHub account is not seven days old. That's not a very big hurdle, however, to uploading stuff to clawhub. So it's a pretty low speed bump, even though I did run face first into it because I had just made my GitHub account. So.

A (47:42)

And this is where at some point, I'm sure the founder of this open source project is going to announce a huge $50 million round of funding at a, I would say $300 million valuation. So 50 million for like whatever, it winds up being 20% of the company that's coming any day now. When that happens, you will see a hosted version of this and I saw Cloud is actually hosting openclaw now and you'll start to see some security precautions be put in place. Yeah, Lukas.

D (48:14)

Yes. And one thing I will add is we've seen and I've spoken to people that work at very big tech firms or tech companies that have similar internal tools that were built out. And this goes back to kind of the innovators dilemma that Google had gone through with their Google DeepMind project and how it took them forever to launch something because they want to satisfy their current customers and also not act too much like a startup. But I think that the large tech firms now are going to start getting involved.

B (48:48)

Clearly, people want more capacity to do stuff on their computers and they don't want to have their hands held. Now, I wouldn't have my parents use this, Jason, but I think for us, us nerds, I don't know why we shouldn't have access to, you know, to our own castle and our own keys. It just seems very reasonable to me.

A (49:03)

Yeah, I mean, nothing can stop you from doing this yourself. Obviously, if you do install this yourself and you authenticate your own accounts, I think you're crazy. But I think if you create a stack of accounts that are read only for this, the notion scale is the other one we've installed. Yeah, Lukas. Yes.

D (49:23)

And we've given it very specific access. And it's all kind of created around multiple different API keys that we've locked down.

A (49:32)

The notion skill is by notion or it's just by somebody who wrote it.

D (49:37)

This particular skill is kind of custom done on our side. So we made our download a skill to correct.

A (49:45)

Okay, so we've got polymarket Reddit, 11 legs gamma last 30 days and then you did the task manager. So that's six. We did six definitive skills today. We didn't get to 10. There's more coming. Any Open Claw news that we should be aware of, Alex, as we get ready to wrap here on day, day.

B (50:08)

Eight of Open Claw mania, a couple of small things. One wiz, the cyber security company that Google purchased for. Oh, Jason, back me up here. 30. 32 billion.

A (50:18)

30 billion plus.

B (50:18)

Yeah, 30 billion plus. You know, they turned down the money last year, blah, blah, blah. They found that multiple book, if you call it Mult Bot malt book. The idea was your agents would have their own social network and talk amongst each other, incredibly insecure, and was leaking stuff all over the place due to a misconfigured Supabase database. So just another warning that when you're playing with fire, you know, don't have your house involved with it. And then there's a cryptocurrency problems with other skills. Again, stay safe. But I just want to throw in a little note to everyone out there listening to this who's kind of learning but not playing. You should go make your own skills. And it's really, really doable if you have multiple AI platforms. I just, whenever I ran into a problem, Jason, I just screenshotted the issue, threw it into my unlimited ChatGPT account, and then used that as a way to unstick me elsewhere in cloud code or other places. And it made me feel like a super, super person. Like it was incredibly cool to build my own software, ship it to myself and run it myself. So if you're listening to this, you're not a founder, not hyper technical. Just give it a try. It's like 20 bucks in credits and you'll learn so much and you'll feel awesome and you'll feel in the know. So just go get your feet wet. It's really worth it.

C (51:23)

Yeah. And Jason, I have a quick question. Peter Steinberger, who was the founder and creator of openclaw, he's kind of seems like a serial open source guy. He's. He's released, I think, around 50 tools over the past couple of years on GitHub, and I can't imagine what life is like for him at the moment. Are VCs just throwing money at him? I think that he made around $100 million from a previous venture of his. So I guess what would you do? If you're in his shoes, what do you think life is like for him at the moment? I've seen him on a bunch of podcasts, so he's obviously enjoying life.

A (52:01)

Yeah, you will get a lot of excitement because when something is this weird and catches fire, it's very rare in our industry. This would be lightning in a bolt bottle. So when lightning in a bottle happens, VCs show up and they love to take risk. What he's going to be drowning in is, you know, the need to ship product, the press asking him to talk, customer support, bug fixes, and raising money. These are all going to be pulling him in different directions. This is why multiple co founders is what Paul Graham insisted on at yc. What we do and we copied from Paul Graham was, hey, we need multiples. It's because when something like this happens, you want somebody to be able to watch the store as the other person goes and gets the investment. Right. And another person deals with the press and the, you know, those issues. So he's probably dealing with that. What he needs to do is make a decision. Who does he want to have as his first investor who is somebody of high integrity, high track record, it doesn't matter at this moment. And you should have good counsel, right? So you should have a great attorney from whichever top tier firm. And I'm sure those firms have contacted him as well. So just like if you were a basketball recruit or an actor or a comedian, you'd want to have good representation, good lawyer. In our world, it would be an agent in the Hollywood metaphor. And then you have to make an economic decision. Well, how do you make an economic decision If I assume 50 people have made him an offer? You just talk to five other founders and say, who's the best investor you ever had? Boom. And he's going to be able to skip going to an accelerator or even founder university. Obviously he's going to be able to go right to a series A. And so he should say, I want to do a series. He should tell everybody, these are the terms. I am raising 20 million for 10% of the company. You get a board seat. There's going to be five board seats. I have four. You have one or one's independent. Here's my attorney. You can talk to them. You know, you get this set of rights. And then he should look Sequoia Kraft, Andreessen Horowitz, gc. Which of those big firms he wants to work with. And then a step down from there, Khosla, who he wants to work with. Who does he vibe with? Does he vibe with Jason Calacanis? Does he vibe with Sequoia Ruloff? Does he vibe with David Sachs? Or David, you know, probably would be somebody at David Sachs firm, now that he's at the White House. Most of the time you would pick who you vibe with and then you'd have a conversation and you just narrow it down. Now what's going to happen is then people are going to outbid. So the people with low reputation would want to outbid the other people. So they say, well, I'll give you 30 million for 20% of the company. I'll give you 50 million for 20%. And that can get a bit, you know, that can get a bit crazy. What you want to do though is avoid just taking the biggest number and you want to go with the best reputation. Best reputation, most thoughtful person who's been in the industry the longest. The other possibility is to say, I'm raising $20 million for 10% of the company and I'm going to do a party round. You can take up to $4 million and I'm going to have five different firms put in $4 million each. And I'm building a bit of a brain trust. That's. Some people might not like that, but they might make the bet anyway because there's so much money in the world and this is lightning in a bottle. Larry and Sergey did that with Kleiner Perkins and Sequoia. They said, we're famously going to give you half each. They were both, they both said, no, no, no, we want the whole thing or nothing. And they said they called their bluff and sure enough, they split the baby.

C (55:50)

He did create a previous company called PSPDF Kit, which is a PDF technology company. And so he's been a founder before. He might know a little bit of the show, kind of what it takes, kind of what he needs to do. He might already have some money in his back pocket for a project like this. Is there a world where he doesn't even need to raise?

A (56:12)

No, there's no world in which he should not raise.

B (56:14)

That's interesting.

A (56:15)

This project is so great. He should just take the money. It's a once in a lifetime opportunity and if he doesn't, it would be a giant mistake. This would be like creating the IP for Superman or the IP for Star wars and not going for it. You know, you're George Lucas, you've got the script for Star wars, you've got some treatments, and Fox says, you know, 20th Century Fox says, we'll produce it, you know, we'll, we'll distribute it, we'll give you the money for it. And then, hey, you've got all the cards because there's so much competition. And he said, yeah, I'll do it with you. I get the sequels and I get the merchandise. And that's what Lucas was able because of how desirable Star wars was and the sequels. He said, I just want to own the sequels, I want to own the merchandising. And then we all know how that worked out. The Star wars toys made more money than the movies ultimately. And so there you have it, and the sequels. Then he got to renegotiate and negotiate over and over and over again for. And he funded themselves. So you take more risks. But you gotta, gotta, gotta close around this week, next week. You gotta have a great lawyer and you have to, you gotta just tell them what you're looking for and why and then call it a day. I would, this is what I would do if I was him. I would pick two top firms, give them 40% each, I take 20%. And then I would go to all the Top influential people. I'm talking my own book here, but people like me or whatever and say, hey, I have 500k for you. And you get, you know, if there was $4 million, you get eight people to put in 500k or 16 to put in 250k. And then you can always call them or email them and say, hey jcal, you know, hey, you're an investor in this. Can I come on this week in startups or can I get a shout out on all in or can you come speak at my Clawbot conference? You know, whatever it happens to be, that's, that would be the high art is to have two firms, two board seats. They each get a board seat, you get the other four board seats, you have super voting or the voting shares and then yeah, you go from there.

B (58:16)

So when we often see open source projects, Jason, the company built around them is often offering a hosted version with, you know, enterprise security and so forth. You also mentioned the App Store idea and that's proved to be a very lucrative method of monetization for Apple and Google. What do you think is the best way for the people behind Open Claudette to monetize this product? So I'm kind of torn about which way I would go if I was them.

A (58:37)

You go, you do both hosted and pro. Then I would do a services company to help people implement it. And then I would do the App Store where they're, these skills are subscriptions. So last 30 days is, you know, X amount per month, per year and you just pay for it as you go. And they take 30% super easy. And for that 30% or 20%, whatever they choose, they have a dev team and security team that reviews the code and make sure it's on the up and up and easy to install on the hosted version. And then they could do a deal for the hosted version version and a licensed version for cloudflare for, you know, with support and everything. So it's usually the support, the services, the App Store, the updates is what people will pay for.

B (59:23)

Yeah. Cause if you told me that I had to pay for the last 30 days, I would just go make it myself and then run it myself. I think, I think now that I have the capability to create my own software on demand with funny little jokes with cloud code, my interest in paying for software has gone down dramatically.

A (59:36)

People can and do fork WordPress or install their own version of WordPress and then there are people who use hosted WordPress. It's really that simple. You make the decision for some people it's well worth the money. For other people, it's not. Easy peasy lemon squeezy. And here's your Poly Market Fun. Poly Market. Show them the Poly Market, Alex.

B (59:54)

It's pretty good. All right, so the name here is a little bit dated, but roll with me, everybody. Today's Poly Market is, quote, multiple AI agent sues a human by February 28th or by the end of this month. And the reason I picked it, Jason, is it has an 83% chance according to polymarket sharps, and nearly $300,000 in volume. So a good amount of volume. And they really do think that someone's AI agent from OpenClaw is going to sue a human. And I would take the no on this, but I wanted to get your take on who's right here. And we'll talk about the rules after that.

A (60:26)

My understanding is a lot of these agents that are talking to each other on Malt Book, which is like Facebook for the agents are being prompted and put up to like they're their conversations. It's not just organic stuff, but sure, somebody will train their agent to file a lawsuit and it will sue somebody. Yeah.

B (60:47)

Is it ethical to create a skill that has your agent sue someone so that way you can win the bet? Or is that unethical in a predict because you're not predicting.

A (60:57)

We're in uncharted territory here, Alex.

B (60:59)

That's why I'm asking. I mean, because I just. I just realized that I could, say, spin up a polymarket account, drop some USDC in there, bet on yes, and then sue you for fun.

A (61:09)

And then I think the fun thing to do is to connect it to polymarket, put some money into the account, and then have it look for wagers to place. And then just because people have done this already with the API, they're already doing quantitative, you know, betting. But now this opens it up to everybody so you could come up with all kinds of ideas of how to bet on markets. I want to bet all the long shots and say they're not going to happen because they're long shots and just make very small amounts of money. Tons of different ways to do it. All right, listen, I don't know what to tell you folks. This thing is taking over the industry. I predict it's not going to stop. I think this is going to keep going. I was just thinking, I have this really great domain name, begin.com and maybe that would be a good way to do a hosted version of Be this. I don't know if there's a founder out there who wants to do a hosted version of this or, like, make a verticalized version of it for small to midsize enterprises. But pitch me on that. We'll incubate it. And I have the. The best domain name ever. Begin.com. all right, we'll see you all next time on Twist. Bye Bye.