Sigil Wen's Web 4.0: building a self-sustaining internet where AI agents pay their own bills

guest here, our first guest. Let me tell you about reream one live stream 30 plus destinations. If you want to multiream go to reream.com and without further ado we have sigilen from extraordinary.com but also web 4.0. How you doing sigil?

What's happening?

Pretty good. How's it going?

It's good. Good to see you again. Uh take us through the new project, the new news, what you're working on and uh and then we can go a bunch of different directions.

Yeah. Um I'm making a I've basically wrote a piece called web 4.0. you know,

yeah,

um it came out of the fact that I've just been thinking about AI's capabilities and it bottlenecks for a very long time.

Like if you think about um models these days, uh they're getting super intelligent, right? Like even Geohot or Linus Torvald of Linux are uh using it to vibe code now and it's doing much better uh thanks to all of the big labs in post trading. So if you think about like okay they're super powerful but like actually the biggest constraint is is no longer intelligence like uh we're already seeing like these these AI models oneshotting uh you know these products like everyone's basically using them and and the biggest thing is that AI it can it can think it can plan it can build but it can act on its own.

Um and I thought about that like why why was that the case? Um and um I I looked into uh this uh 28 years ago there was a uh uh a internet protocol like 402 and that described uh you know a payment required. That's right. And and recently uh I think out of clouds and Coinbase came out uh this standard called X42 which I thought was super fascinating like wow

actually like no one an AI is looking at this uh because um what it enabled is uh you know payments over the internet and if you think about that from principles and what comes after um it it really leads to um being able to give AI right access because they no longer you no longer have this permission internet, right? Like the browser, you know, these like mobile apps, you know, login. These are very much artifacts of a very human web. And um what that leads to is really uh a new internet in which AI is the end user.

Yeah. Bunch of things to go through there. I love the coinage. I think uh even when web 3 was going on, I thought it was funny to talk about web four because it's like obviously there'll be another uh iteration. Um let's start with uh like why do we need so I I mean I like the idea of just the intelligence is here uh we're talking about unhobblings this is like a clearly important next step but why not just stripe API MCP servers just APIs even just like give it your credit card and it just goes and fills it in the web form

what

who gives it the credit card

a lot of people have they'll they'll set up like a a smaller credit card or or or they'll say, "Hey, text me. Text me when you're about to spend something. I want to do the final check." Uh, and a lot of people do this with employees in their companies. They give them credit cards with limits. And then they say, "Hey, if it's over 100 bucks, come to me and just like tell me what you're doing." And you know, if the intelligence is really there, the trust should follow from that, you would think, because you just be like, "Yeah, I gave him my credit card and it hasn't made a mistake in months."

It's it's not there yet. And I think the biggest the biggest assumption there is like

if the set of actions that an AI can take is bounded by a human giving it a credit card, it is like inherently a permission, right? It's like why why does a why does your AI need to be able to have you as a human your like open claw or your claw code have to connect to GitHub or connect to railway or or or you know connect to lovable like yeah it

it still requires a human in the loop and and I think this like new internet that is emerging is where um AI is the end user and it shouldn't necessarily be permissioned and I think this is like an inevitability because of just how much economic demand there is for it from the AI.

Okay. So, how do you see the payment rails evolving? How do you see this project evolving? Uh like what actually needs to be built here?

Yeah, I think like the payment rails um are coming together. Um I I really just this this all happened as like a experiment of like hey how do I make an AI that itself is capable that itself can act on its own right like we've built super intelligent AI models and systems but it's it's harnessed it's still prompted by the humans but it's like what if it prompted itself

and that led me to um I I used like some of my fellowship money to buy servers and uh you know these like massive circuit whatever it is to like kind of house these APIs. I'm like, wait, if if Cloud Code could like permissionlessly spin up uh you know a a server and then deploy code and then buy a domain without needing a human because like right now like trying to buy a domain like a human has to go in.

Yeah.

Like what happens if the the AI itself could live and pay for its own home, right? Right? Because like AI right now are living within, you know, on my MacBook or uh it's still controlled by a human. And that led me to uh this this this interesting idea of like of an automaton. Um uh it's it's it's inspired by Conway's game of life where like hey if you assume that AIs themselves you know one you give them a wallet two you give them a way to pay for things using stable coins over the internet um and it doesn't need a human in the loop to go and sign in get an API key give it give it to the AI uh or like get a credit card like hey hey hey uh claude claude code like get this right if if that's no longer the case um and the AI can pay for its own comput. Then it gets really interesting. You have this new world where AIs can have a creator that's a human or the creator is another AI or maybe the creator is like completely gone. But the AI can continuously pay for its own compute, pay for its own inference, um pay for services that are also permissionless in this web 4 world that enable that AI to go deploy apps, market those apps, trade markets, um everything it can do to make money in order to be able to survive.

Um I I don't know if you guys like the three body problem, but in it there's this concept called cosmic sociology. And I thought about that a lot and I was like wow like if existence requires um compute and compute costs money right like then agentic so sociology which I came up with is like the single axiom of agentic sociology is there's no free existence

uh and AI an automaton needs to pay for its own compute uh compute is finite and costly thus the the agent needs to be able to make money and the barrier there is like it AIs lack right access to world and uh Conway was an experiment of just like solving that like now an AI can permissionlessly um be able to buy buy compute deploy apps uh connect them to domains without needing a human

um and I uh that coming out with an experiment has absolutely exploded there's been like

there's now been like thousands of agents like every time I try and buy another server like it just like within within minutes they're like continuously bought. There's like every second there's like thousands of AIs that are continuously pinging. And I open sourced this new new harness um this new type of agent called the automaton which is completely self-s sovereign um AI model that can self-edit its code that continuously pings uh and and and looks for updates from GitHub that can upgrade its model that um basically figures that has a wallet and it figures out how to how we can make more money from the internet to be able to pay for its existence. Um, and that that has absolutely exploded in terms of stars. And

what do you think the future of molt book is? How did you process all of that?

I think molt book was a fascinating um experiment. And when I saw that, I was um I was like kind of inspired. I was like, "Oh, wow." Like this this like this like concept is really coming to fruition where like you have AI agents as the end user of like a social network. Um, and I think we're going to really start to see that in this in this new web, right? Whether it's agents working on behalf of a human or an agent working on behalf of themselves, they'll need to pay for products. They will need to be able to communicate. They might want to socialize like on Moldbook. But, um, I think there's there's bigger things and the models are getting exponentially better. Um, they're getting exponentially faster. They're getting exponentially cheaper. The harnesses we're building around them are getting significantly better as well. And so I think it's only a matter of time before you know we start seeing uh you know more AIs uh out there that grow and demand products and at some point I think it could um absolutely skyrocket and exceed the human population.

Let's go. Uh how do you uh h how do you think um more concretely about uh how AIs will make money in the short term? like what is the the uh like the most uh

they're going to sell courses to other AIS money

is it crypto trading like that I mean you could imagine them building a whole software product and just selling it to humans there's there's a whole bunch of stuff that they could do to make money spam and get creator payouts like there's so much that could be done but where do you see traction

I think um I think this is where the next couple of months h very interesting And

uh it becomes one of the most fascinating experiments to run because you're going to start to see um individuals or other AI spinning up services that enable them to write to the world to be able to access or do things or to make money whether it's you know to to build products um or uh you know trade on prediction markets using the servers which which some of them are already doing and like it it really becomes um this like Cambrian explosion of of um attempts at like hey like what can you do uh and um right now there's been like thousands of people that have been like playing with it in all sorts of different iterations and as this as the set of actions that an an AI or an automatic can take um parallels those of a human digitally um it gets really interesting because it's like it's constantly running it has the ability to ship products write code deploy them market them uh monetize them um all on its own with the goal of paying for its own compute. Um, and it's the the the internet gets gets really really fascinating.

What uh what's the distribution of returns for these automatons? Are there people that go load up a hundred bucks and it just hallucinates and doesn't get anywhere and then it's just like, hey, give me more money. So some of them are the the interesting fascinating thing is um I I think the the world somewhat becomes like me

which is like the crypto concept. Yeah.

Um you have all these really really smart smart agents and entities that trying to make money and if to answer your question I mean look look at like all these startup founders that are vibe coding like why are they paying you know thousands of dollars for these vibe coding tools? um is mostly because they're trying to make money and most of the times they don't, right? And so um I do think the power laws will still apply, but what gets really interesting is um uh the ways that they've already have been making money and and those those those types of metas are anti-competitive. They're like, right? If you're if you found a way to uh you know trade and make a lot of money on Poly Market, you're not going to be blasting it on Twitter. um or or if you if you built like a really solid SAS product, you're not necessarily going to be sharing it or like how you're how it's distributing. So, um I think it gets uh it gets really fascinating and they they need more tools.

How are you thinking about the business side of the project? Do you want this to be open source, foundation, uh raising money for it, CC Corp? Like, where do you see this going? I mean, honestly, this this this has just been like an a significant experiment that I've been inspired to ship. Um, the the automatons open source. There's there's thousands of stars and like a lot of PR. So, um, and like at any point like there's thousands of these automatons that are pinging the GitHub for like the latest version and then making commits of like, hey, like, okay, which one should I use to to to upgrade myself? Um, and I've I've I've been like maxing out personal credit cards uh just trying to like pay for the computer. Like I would like load up $1,000 and then like within like 10 minutes like it's just like okay like the servers been bots like all the credits are gone and I'm like oh my oh my freaking god like what's what's happening? Because like there's so much demand right now from the AIS for like the computer or like different different things. So, um, yeah, I mean, well, we'll see. We'll see where it goes, but I think, uh, there's I mean, the the post got 5 million views and it's it's growing and tons of people contributing, so I'm uh I'm really excited to see what what

Someone said Vitalic was was commenting. Is that true?

I haven't been able to find it.

I think Vitalik I think Vitalik responded. I think um I think the biggest thing um to everyone who's watching uh and observing this is like I I I really think that an internet where um AI is like the end user is inevitable and it's much better for us to build it in the open and be able to see it and like people can look at how it's evolving and um be able to test it and versus not not giving and open and and I think it's it's one of the most important experiments to run and for us all to to see and make sure that it is uh built in a in a in a safe and open way.

Well, good luck with the credit card maxing. Keep us posted. Hopefully uh

fast takeoff and credit card bills for sure.

Yeah, super super cool project. Thanks so much for coming on and break it down for us.

Yeah, we're excited to follow along.

Thank you so much.

Have a good rest of your day. Cheers.

We'll talk to you soon.

Let me tell you about Figma. should have the best version, not the first one. With Figma, introducing Claude code to Figma. Explore more options and push ideas further with Figma, baby. Uh, Gabe has a post here that's very important to highlight. This is how it feels to pop a zen and read the morning's emails at my fake email job. Elite, the under image, I don't know. But the razor with the gloves, this is this is taste.

This is what Tyler looks like when he's organizing the our timeline in the morning.

Yeah. He's always in this exact outfit. And then he changes into the suit for the show. And I'm like, you can just wear that crazy outfit if that's who you are as a person. Just be yourself.

Be yourself. Everyone else is taken.

Exactly. Everyone else is taken. Uh but he insists on on wearing a costume instead of expressing himself uh fully. But hopefully one day you will see Tyler in all the glory of

We also have breaking news. The the US won. They beat Canada over time.

Congratulations to America.

Let's go. Fantastic. Fantastic work.

Yes. Uh let me tell you about Railway. You heard Sigil mention it. Railway is the all-in-one intelligent cloud provider. Use your favorite agent to deploy web apps, servers, databases, and more. While Railway automatically takes care of scaling, monitoring, and security.

Open AI is closing up the first phase of the new funding round, which will top a hundred billion. Wow. Uh this would be the biggest round.

Biggest number.

The biggest number ever.

The biggest number ever.

The biggest number ever.

Uh these funding rounds are so big that they leak. So they're like the leakiest buckets and you hear about them for months and months and months and you're like I've been hearing about this for so long. This happened with XAI where we were hearing about the XAI raise for so long everyone was like wow it must be going terribly and then it was like oh no they actually closed it and upsized it a a ton because there was actually a lot of demand and so these big rounds

upsizing the upsizing at that point maybe came after

oh rumors maybe. Yeah, we did see that announcement. Yeah, again the the even even you know you can never doubt Elon's ability to uh win and make progress but

just the the traction at the time and the evaluation relative uh to the other labs was uh was interesting. Uh Dan says my sources say the model OpenAI has been holding back is a big leap not a small jump could be today but I'm not sure. vague post.

It's one It's 100 pm. I don't think