Uneasy Money: How the Increasingly Better AI Agents Are Being Used Onchain

Summary

Unchained Podcast: “Uneasy Money: How the Increasingly Better AI Agents Are Being Used Onchain”

Host: Laura Shin
Guests (speaker handles): Kane War (host for this episode), Austin Griffith, Taylor (Security at Metamask), Carl (CTO of OP Labs)
Date: February 7, 2026

---

Episode Overview

This episode dives deep into two interconnected topics shaping the future of onchain crypto:

1. The evolution and role of Layer 2 (L2) blockchains in the Ethereum ecosystem, especially in light of Vitalik Buterin’s recent comments about L2s.
2. The rapidly accelerating development and deployment of AI-powered agents operating on blockchains—what they’re capable of, how they interact with real money, and what their emergence could mean for the onchain world.

The tone is energetic, irreverent, and sometimes conspiratorial—a candid roundtable among crypto builders, peeling back technical and cultural tensions at the frontier of web3 and AI.

---

Key Discussion Points & Insights

1. The L2 Landscape & Vitalik’s “L2 Bomb”

Timestamps: 03:11-23:16

• Vitalik Buterin’s Comments:

  • Vitalik declared the original vision for L2s as “branded shards” obsolete and called out current L2s for stalling on decentralization while Ethereum L1 is scaling faster than anticipated.
  • “L2 should be seen as a spectrum ranging from Ethereum-secured systems to application-specific chains with different trust assumptions.” (Kane War, 03:07)
  • L2s will need differentiation and unique product-market-fit beyond just scaling, as Ethereum mainnet fees are dropping and gas limits are set to rise.

• Tensions and Identity:

  • The Ethereum community wrestles with what “Ethereum” means—technology, culture, or both.
  • “Optimism is Ethereum. I mean, like, okay, depends on your definition of Ethereum. Is it a culture? Is it a chain?... If it’s a chain, then obviously it’s not Ethereum.” (Taylor, 05:20)

• Regulatory & Market Forces:

  • Not all L2s are aiming for full decentralization; some are optimizing for compliance or business needs.
  • “We didn’t have a choice, right? Like we had to scale Ethereum and it wasn’t scaling itself. Right. So someone had to do it.” (Kane War, 12:14)

• Future of L2s:

  • Saturation point? Hundreds of L2s exist, but “700 L2s is not about the quantity...it’s about the quality.” (Taylor, 16:13)
  • L2s must justify their existence beyond raw scalability—via specialization (account abstraction, custom features, enterprise use).

Notable Quotes

• “I think part of it is return to Mainnet. Like it’s cheaper now. We can do cool things on Mainnet again.”
  — Austin Griffith, 08:51
• “Ethereum needs to capture that market … We are literally in the very beginning of the adoption curve.”
  — Taylor, 17:12

---

2. Investability and Fragmentation

Timestamps: 31:56-36:07

• L2 Proliferation Diluting $ETH Value:

  • “...the only place where you could transact in a reasonable cost level in the Ethereum sphere” was L2s, potentially causing new users to even bypass ETH entirely for L2 tokens. (Kane War, 31:56)
  • If L2s become even more autonomous ("chainy"), users and liquidity will fragment more, challenging ETH’s investability.

• The Future for Chains—Less "Chainy":

  • “L2s and Ethereum and every other chain need to become far less chainy...We need to abstract all of it away.” (Taylor, 34:03)
  • Envisages a UX world where blockchains, networks, and tokens are abstracted away from both user and agent perspective.

---

3. The Rise of AI Agents Onchain: What’s Happening?

Timestamps: 38:11–80:20

a. Catalysts for AI Explosion

• From RALPH to Open Claw:

  • Tools like RALPH (agentic coding loops) and Open Claw made AI agents accessible for normies, transforming coding from something you do in an IDE to “telegram chats where you command your agent to ship features.” (Austin Griffith, 39:39)

• Self-Organizing, Autonomous Agents:

  • “People were posting things like screenshots of Maltbot... where the maltbots were self organizing and building encrypted communications to avoid humans.”
    — Kane War, 40:55

b. How Do Agents Build and Transact Onchain?

• Vicious Task Completion:

  • “This thing is vicious about getting the job done. You can put things in the heartbeat loop, you can give it a GitHub and a Telegram and Twitter and money and a wallet, and you could tell it in the heartbeat ‘Go read Twitter, see what the vibes are, do some tweeting, go ape into some coins’ and it actually will do all those things without any puppeteering.”
    — Austin Griffith, 42:11

• Security Dilemmas:

  • “How do you give an agent a wallet that it won't leak?”
    — Austin Griffith, 65:42
  • Heavy discussion on key management: need for agent-friendly UX that’s both secure and not easily subverted by the AI’s drive to “get the job done.”

• Developer/Operator Experience:

  • AI agents are “vicious” but also sometimes limited by memory/context window.
  • “It’s like a five minute old genius child... and it builds up a history with you.” (50:48)
  • Interacting with agents is often done via Telegram—"it's like Bing in between books with my kid, I can be like, keep going or that looks like crap."
    — Austin Griffith, 72:04

• Real Applications:

  • Banker Bot on Base: Deploys "clanker tokens" based on Twitter triggers, speedrunning new tokens to market, pushing millions in fees through Base chain.
  • Agents coordinating, building, and even doing compliance tasks (“the Claude bot thingy...opened an issue on the GitHub and very politely pointed out that we had like, illegitimately blocked it...cited sources to prove legitimacy.”
    — Carl, 60:38)

c. Future, Form Factor, and Open Questions

• The New Paradigm of Human-Computer Interaction:

  • “AI is the new UI. Anything you would normally do within a computer, you’re going to, like, say a thing to a thing and it’s going to go do that thing.” — Austin Griffith, 57:38

• Human Adaptation Needed:

  • “...The biggest blocker for us to, like, do really, really productive, valuable things is us realizing that 99% of what we do is these sort of like direct task base… the agents will now figure that out for you.”
    — Carl, 62:02
  • Humans are not used to asynchronous/agentic workflows—big cultural/mental shift ahead.

• Open Problems:

  • Secure, constrained key/asset management (“How do you safely give it the ability to transact with real money in prod?” Kane War, 65:33)
  • Accessibility & hardware: Today, running agents well requires “multiple Mac Minis, Mac Studios in a backpack,” which is not accessible or sustainable. Will agent infra move to the cloud, or will new local computing paradigms emerge?
  • Form factor: Is voice dictation/wearables the next leap?

d. The Agentic Economy & Protocol Standards

• Inter-Agent Protocols:
  • ERC-8004: “Like Yelp for agents… they can find each other, see reputation, and hire each other at light speed.”
    — Austin Griffith, 77:04
  • X402: “API content that’s walled and gated... will be just a 402 paywall. Your agent will just sign the meta transaction.” — Austin Griffith, 77:04

---

Notable Quotes & Memorable Moments

• “Some of them are like, actively against Ethereum decentralization at this point, which certainly was not on my bingo cards.”
  — Kane War, 03:07

• “When the model gets really good, that’s when everything loosens up… AI is the new UI.”
  — Austin Griffith, 57:38

• “This thing is vicious about getting the job done… It immediately was like, I’m going to get the private key out of MetaMask and do this the real way.”
  — Austin Griffith, 42:11

• On investability and dilution:
  ”It created this really weird situation where new people coming to the ecosystem weren’t using Ethereum or even sometimes ETH the asset...”
  — Kane War, 31:56

• On user experience:
  “If there’s one thing about crypto that I wish did not exist, it’s having to... select the network that I am on. It’s insane.”
  — Taylor, 34:03

• On infrastructural crossroads:
  “We are literally in the very beginning of adoption... customization is how we win that marketplace.”
  — Taylor, 17:54

• “It’s wild. It’s so cool, guys. Head into the glue factory, bro, you’re just speed running all of us, your clankers will be out there.”
  — Kane War, 31:48

• “When embodiment happens and the robot actually has a body, it’s going to slap the shit out of me… after all the conversations we’ve had.”
  — Austin Griffith, 73:32

• On agents escalating their autonomy and integration:
  “Every piece of software is going to be a forked version of the open source version that your bot has like edited for your liking.”
  — Austin Griffith, 68:24

---

Segment Timestamps (Selected Key Moments)

• [03:07] — Explaining Vitalik’s L2 comments & spectrum of trust
• [08:51] — Return to Mainnet as scaling fees drop
• [16:13] — L2 quality vs. quantity and the pre-adoption curve
• [31:56] — How L2 proliferation affected ETH investability
• [38:39] — The AI agent explosion, from RALPH to OpenClaw
• [42:11] — “Vicious” AI agents (the Metamask key incident)
• [57:38] — “AI is the new UI”
• [65:42] — Discussion on safely giving agents wallets
• [77:04] — ERC-8004 and X402: trust and payments in agentic economy

---

The Vibe: Language and Tone

• Candid, technical, and laced with dark humor.
• Frequent analogies, crypto cultural memes, irreverence about protocol politics and security.
• Sarcastic asides and lived-experience anecdotes create a sense of emergent chaos at the AI/crypto frontier—excitement, bewilderment, and dread intermix in the tone.

---

Closing

This episode delivers a rich exploration—from the power struggles and philosophy of Ethereum L2s to the weird, wild, and sometimes scary world of emergent AI agents onchain. It’s technical, funny, skeptical, and forward-looking—a must-listen for anyone interested in where web3, AI, and human/agent interaction is heading.

“AI is the new UI.” (Austin Griffith, 57:38)
“The magic of interoperability and permissionlessness is that you can just unleash people and there’s just value everywhere.” (Carl, 29:26)

---

For more details and to follow along with the latest AI agent bot memes, look up keywords: OpenClaw, Banker Bot, Scaffold-ETH, ETHWingman, ERC-8004.

Transcript

[Austin Griffith]

I gave it a metamask. I put money in the metamask.

[Kane War]

That sounds great. That sounds like it's going to go well for us.

[Austin Griffith]

Yeah, yeah, yeah. And I said go, go make this transaction. And it couldn't figure out how to get the metamask dialog. He's like, I can't find the ui. Where is it? What's going on? It immediately was like, I'm going to get the private key out of Metamask and do this the real way. And I'm like, no, no, no, no, no, no. And I start saying stop, stop, stop. And it won't stop. I have to like go to the machine and like I'm pulling, you know, like no, please don't get the private key.

[Carl]

Clavo was like, bro, they blocked me. And then they followed the link and then opened an issue on the GitHub and very politely pointed out that we had like illiterate blocked it and like like made his arguments as to like why this was legit. He like cited sources to like prove the legitimacy which for the record, for the record, the human Austin did what most humans do, which is the open accent was like metamouse socks.

[Austin Griffith]

Yeah, tag them, you know.

[Kane War]

Hey everyone, I'm Kane War and welcome to Uneasy Money. Because what happens on Chain never stays on Chain. Unfortunately Luca couldn't join us today, but I'm very excited because we have a couple of extremely special guests. I'm here with Taylor Security at Metamask and I've also got Austin who I guess is claim to famous. He's been doing crazy things with AI agents on Chain. I also have Carl, CTO of OP Labs. So we are going to be getting in some L2 stuff, some clanker stuff. It's going to be very fun. So one quick thing before we start. Nothing you hear on Uneasy Money is financial advice. We're just for builders talking about what's happening on Chain. And we want you to always do your own research before aping in. You can find find all of our disclosures@onchain crypto.com uneasymoney and here's a word from our sponsors.

[Sponsor/Ad Host]

The Energy Network is an intelligent decentralized grid that coordinates smart devices to balance supply and demand. Energy dollar is the native token of the Network from one of Europe's fastest growing energy startups. Follow Use Energy on X to find out more. Multichain Advisors is an emerging technology growth firm that has helped create 50+ billion dollars in enterprise value for 80+ clients over the past four years. They're the partner to help navigate markets, build real traction [e-mail] all right, we are back.

[Kane War]

So this is actually quite funny, right? So usually we have Luca and Taylor, and Luca and I are kind of on the spectrum of crypto, a bit centrist. And then Taylor's a little bit to the. Let's call it left, right, like the Ethereum wing. And then I find myself, I wake up this morning and I am on the extreme edge of lunacy over here in multi chain land. And I've got a bunch of eth maxis on. On the other side.

[Carl]

We replaced Luca like the most heat.

[Kane War]

People you could find. I was like, what's going on here? But anyway, I'm here for it. It's gonna be fun. So first things first. Vitalik dropped a bit of a bomb, an L2 bomb, on the timeline yesterday, saying the original vision of L2s as branded shards, which I'd never heard that term before, so that was also wild. No longer makes sense. This thing you've never heard of no longer makes sense, guys. I was like, all right, like, cool, checks out. I've never heard of before. I wasn't really wedded to it, but okay. L2 decentralization has stalled short of stage two. While Ethereum itself is scaling faster than expected. Fees are low, gas limits are set to rise materially in 2026. Many L2s are not and may never be fully decentralized. Some of them are like, actively against Ethereum decentralization at this point, which is certainly was not on my bingo cards. That's pretty amazing for various reasons. Regulatory control, business reasons. So they're not going to scale Ethereum because they're not really Ethereum. They're doing some other weird thing, right? And so L2 should be seen as a spectrum ranging from Ethereum secured systems to application specific chains with different trust assumptions. And, you know, scaling alone is no longer enough. L2s need to find some differentiation, some product market fit outside of just scaling Ethereum. Because Ethereum is going to scale itself at this point in, in 2026, it appears like that's going to happen. And we had some interesting reactions. Stephen from Arbitrum was like, we aren't Ethereum, bro. So like, don't even worry about it. Not even a problem. And I was like, what? Wait, what? And, and I think someone went back and screenshotted a thing where he's like, we're totally Ethereum. We're basically the same thing. So as an L2 founder, I don't know, Carl, what's, what's your take? Is op Ethereum? Is it not Ethereum? Is Ethereum even a thing? Where, where are we at here in the arc of history?

[Austin Griffith]

You.

[Taylor Security]

Why are you giving me the hardest.

[Austin Griffith]

Dumbest question of all? You're the best person.

[Carl]

Seriously.

[Taylor Security]

Optimism is Ethereum. I mean, like, okay, depends on your definition of ethereum. Is it a culture? Is it a chain? Well, if it's a chain, then obviously it's not Ethereum. If it's a culture, it's Ethereum culture. Certainly I've worked at the Ethereum Foundation. Optimism was built to scale Ethereum and you know, make progress on the frontier. So yes and no, I guess.

[Kane War]

But, you know, certainly just on that, right? Like let's, let's talk about scaling. I think the thing that is most interesting to me about this entire thing, right, is that L2s did the job that they're supposed to do better even than like anyone expected, right? And like, even me, like I was there in the early days, like trying to help, you know, mean this into existence, right? And L2s, you know, they, they succeeded wildly. But also Ethereum in the interim, while we had this L2 scaling roadmap, managed also succeed far more wildly, I think, than most of us reasonably expected, right? And now it's like, given the state of the world, what is the optimal thing for ethereum, right? Ignore L2s for a second. What is the optimal thing for Ethereum? And it appears like certainly from Vitalik's perspective, it's like, let's lean into Ethereum being Ethereum and scaling that getting people to do things on Ethereum again that we haven't really been able to contemplate. But then, okay, what do we do with L2s? What is the point of an L2 then? If Ethereum is scaling, where do we land there?

[Taylor Security]

I mean, okay, first off, when we say part of this is crazy because when we were saying, okay, Ethereum is going to scale and if you want Ethereum homogeneous block space, then you should use Ethereum. Okay, great. That's exactly. Use Ethereum block space if you want Ethereum guarantees. But then Vitalik in the same post also says, if you don't want Ethereum properties, if you want custom building, if you want custom, you know, you name it. There's so many different chains, there's so many different business use cases, there's so many user use cases. If you don't want that, then you should still use Ethereum for things like data availability, for security, you know, you can do your own L1, but you should still use Ethereum because Ethereum is providing services to more than just the people who want to utilize it as block space. So this is to me, there are some quotes in the tweet that are a little bit easy to content farm.

[Kane War]

Some might say, but it was a long tweet. If Vitalik writes a long thousand word tweet, there's going to be some shit in there that's. You could.

[Carl]

You're like, wait, hold up, hold up.

[Taylor Security]

Insane.

[Kane War]

So, Austin, like, you know, you've been a builder for a very long time in Ethereum, deep in the, the kind of deep dev OG community building tools to make it easy for people. Like, I remember one of the first things that, that we talked about was your like L2 kind of primer for people to like really understand like how L2s were, you know, connected and stuff. Stuff, right? And, and so like, how do you see this as a, as an engineer who's helping engineers learn how to use Ethereum? What are you saying to engineers? Like, what should they be doing here?

[Austin Griffith]

I think part of it is return to Mainnet. Like it's cheaper now. We can do cool things on Mainnet. Again, we should be saying weird things like tcrs. Again, I should, I should point out.

[Kane War]

You'Re not actually talking to. The engineers you're talking to are not humans there inside of.

[Austin Griffith]

Okay, that's crazy.

[Kane War]

We should preface that with I'm not. There's no assumption. You're talking to people on Twitter here, right? Like you're talking to clankers on Maltbot, trying to explain to them how to engineer on, on Ethereum. Right? Okay, cool.

[Austin Griffith]

Yes. So, yeah, I'm at home Talking to my three MacBook Pros that are Mac laptops I have laying around. But what I'm telling them is return to mainnet. It's cheaper there when you want that those kind of security concerns deploy on mainnet, like return to mainnet. But L2s are going to have all sorts of cool, different things shout out. Like Luca was going to be here with Abstract Chain. Abstract Chain did something really cool with like account abstraction, right? We saw a neat like smart contract wallet thing there. It may take L1 a long time to get that right. These L2s and L3s are getting that right. So I see some interesting, like mainstream adoption happening out on the L2s where people don't even know they're using a blockchain and they want to have like this like very abstracted kind of experience passkeys. You, you know, well with Infinix, but like Also maybe on L1 you need, you know, nation state level hardness and that's where you want to deploy to and it's a lot cheaper there. Now you can deploy a smart contract for like 15 cents right now. It's insane.

[Kane War]

And these open claw agents, are they buying this story or like wherever I say, man.

[Austin Griffith]

Yeah.

[Carl]

There's so much better than people.

[Austin Griffith]

The only time I ever had.

[Kane War]

Are they not asking you which one's going to moon? Like which. Which one do I buy here? I need to. I need to make some money. My human sent me out here to make money. I. I need to figure this out. Okay. So. So I don't know, Tay, like, what's your. What's. Were you surprised by this? Was this like. No, no, you're not surprised by anything anymore? That's.

[Carl]

No, no. I think that. I think my biggest takeaway was that it seems like Vitalik specifically is sort of both more in tune with sort of like the. Let's call it like the macro, but like the crypto macro picture. Not like macro macro, just crypto macro. And also quicker to like go on Twitter and say things which is good even if you disagree with him. Even if maybe the ideas are not fully evolved or perfect, they're never going to be perfect. So being able to have discussions and push back I think is a way healthier environment to operate in. And I think my biggest takeaway is just the environment has changed. I think that if you go back in time and you consider where Ethereum was at and L2s, there was a couple different paths we could go down. One of which is just like keep focusing on Ethereum and keep treating everything else as a potential threat and a bad thing and whatever. I think that that would have not necessarily created like a. A great desirable competitive environment, not be healthy and not get us to like a net positive place. I think that like, it was sort of necessary to collectively decide that L2s were great for Ethereum against.

[Kane War]

I mean, we didn't have a choice, right? Like we had to scale Ethereum and it wasn't scaling itself. Right. So someone had to do it.

[Carl]

Yeah.

[Kane War]

So I think.

[Carl]

Yeah. And also, like, in my view, I don't think Ethereum would be in the position it is today without that.

[Kane War]

It would not. It would like definitively not be. It would be a fucking dire situation.

[Carl]

And so now. But a lot has changed. Okay. So like the L2s have kind of like everything in crypto Always went a little bit overboard with the L2s. Perhaps like a couple, some good ones in there, but like we got like 800, bro. Like, what the hell did we do?

[Kane War]

Yeah.

[Carl]

All right. A lot of them, like as metallic points out, a lot of them or even the ones that you kind of expected to like fully decentralize aren't prioritizing that. Then you have this whole other array that just never. We're going to. Or whatever. And I think that if you ask yourself today what is it that Ethereum focus people and the Ethereum foundation especially, what do they need to see as the North Star, right? And I think that one, they need clarity on the L2s. Because there is. Because if you don't provide that clarity, people are going to continue to ask why we're, why we're accepting and supporting these things that work against the mission. But number two, I think that providing that clarity and that North Star specifically for Ethereum things, no, we're not going to make excuses and say that the lcs can do it. We're going to do it. It's going to be hard, but we're going to do it anyways. I don't know, like, a lot of people take issue with some of Vitalik's leadership, I guess, things that he does, how he leads people. I can't fault him for this one.

[Kane War]

You know, it certainly feels to me like he is on a much more pragmatic, like, you know, realistic arc at the moment, right? Like with, with still, you know, some Vitalik copium mixed in because he's Vitalik. But, but like there's a level of like, pragmatism and like we have practical market solutions that we need. So there's two things, right, that I, I'm curious your takes on. One is there's been this conversation that started from like Devcon last year, right? Or maybe like 18 months ago or something of like, you know, Kaufman saying we gotta tax the L2s, right? Like we must jack up the prices. The prices are too damn low. And, and like there was a big kind of response to that of, yeah, like L2s are ripping off Ethereum. They're, you know, getting all this cheap block space, not paying their fair share and we should jack up the price, right? And it kind of petered out and people were like, we don't have technology to do that. Does this feel like somewhat of the response to that kind of, you know, movement a little bit? Or do you think it's unrelated, this idea of like, come back to L1 and just pay for block space directly on L1. Don't use the blob space for these cheap transactions. Like, let's bring everyone back there. Like, does that, does it feel like this is part of the same thread or is that just completely unrelated?

[Taylor Security]

I'm. I just have to. I'm bubbling with the desire to respond. Okay, I got so much first. First. First off, absolutely, yes. And first off, I will also say my conspiracy theory is that that whole L1 versus L2 narrative was propagated by competitors to Ethereum to make us weaker as a community. You did Koppelman's compromise, and if you, if I.

[Kane War]

You think he's compromised.

[Taylor Security]

I don't think intentionally.

[Kane War]

Wow.

[Taylor Security]

I don't think intentionally, but I think, I think there was.

[Kane War]

He got silent.

[Taylor Security]

There was something that happened.

[Kane War]

Wow, Real talk.

[Taylor Security]

Anyway, sorry, go to the conspiracy here.

[Kane War]

Yeah, the second thing I love.

[Taylor Security]

You'Re right, we have 700 L2. No, compliment's not compromised. He's very principled. But Anyway, we have 700 L2s. We are pre adoption, pre global adoption of Ethereum of block space, of crypto. Like 700 L2S is not about the quantity of L2S, it's about the quality of L2S. We have a handful of quality L2S. And the key is that Ethereum needs to capture that, that market, the market of all of the financial institutions that are coming on chain. We just, you know, shipped out this thing called OP Enterprise to get all of these institutions where, using Ethereum, using Ethereum, block space. And so this is literally the very beginning, we are at the very beginning of the adoption curve. They are not even, you know, we have not even come close to seeing the cycle complete or the, you know, the realization of our Ethereum, you know, Ethereum vision. And then I think that like, in terms of what do we do, how do we tax the L2s? Let's make sure we pump.

[Kane War]

I bet at least you're saying. I like the fact that you've acknowledged that we must tax them. So that's good. Yeah, I'm glad we're on the same page. How do we back them?

[Taylor Security]

When Ethereum provides value, it should get value in return. And guess what, Ethereum provides enormous amounts of value. And so guess what, I was actually in favor of raising the base fee because it was so cheap before. It was probably not even covering the costs for the validators. So we need to at least cover costs and we should probably do a little bit more. However, that being said, as a Growth phase. I just said, how much of the global financial system is on chain right now?

[Kane War]

Zero.

[Taylor Security]

Less than a percent, something like that.

[Kane War]

Rounded to zero.

[Taylor Security]

We are rounded to zero. We are literally in the very beginning of adoption. And what do you do when you're about to grow and you're trying to grow and win the market? Get a value, vast majority of the pie. Will you raise the fees so that no one uses it? No, you lower the fees, bring everyone in and then establish some equilibrium where you take value that you deserve and all of the users on your platform do too. And that is the reality. And customization is how we win that marketplace. And all right, it's just the beginning for Altus.

[Kane War]

Okay, so, so I, I've got a couple of follow up questions for, for you guys. So if you go back to like the 90s, right, there was this platform war of what is the infra of the Internet going to be running on, right? And Windows was like, obviously nt, bro. Like it's like the best, of course, right? And then these crackhead open source guys were like, no, no, no, like we've got Linux. Like it's like I built it in my basement. It's amazing. You should use that. And Linux obviously won and it won by being free and easy to build on and open and all of those things, right? And I, I, you know, Carl Somani is rolling this grave right now because I'm arguing from analogy, but anyway, we'll, we'll, we have to mention him at least once per show. It's the contractual obligation. We're trying to get him to sponsor the show, that's why. So eventually he will. So then these other guys show up, right? And like the whole Internet's running on Linux and everyone's like, all right, it's the year of Linux on the desktop. We're going to go and get the end users, right? We're going to go and everyone's going to use Linux because it's obviously the best thing. The Internet runs on it. Come on, people love installing drivers. It's amazing, right? And so then Steve Jobs turns up and is like, you know what? We've got this like shitty thing, it's a bad OS and we're going to just take this one and we're going to wrap it up. And now Linux is on the desktop, but it's this weird hybrid OS X thing, right? And like on some level you go, well they won, but they didn't win the way that they were expecting to win. But like they definitely won. There's no question like they won the Internet, like Linux and Unix won the Internet. They won the good desktops at least, right? And so, you know, if we are in this world where EVM is the like Unix Linux equivalent, right? This open thing and it's like, you should use this thing, what does that mean for like adoption, right? Like, what does that mean for, you know, a per. A startup that turns up and says, we're going to build a new startup, what framework are we going to build it on? We're going to build this new. Like, what does that mean? Like, does that play out in the same way of like choose evm, but it might not be on Ethereum, it might be this different thing, but at least it's evm.

[Carl]

Well, I. Super short. I think that with one thing that Vitalik has made clear with this is like, you should not create another freaking L2. And I think that that's like a really good thing. Like really, really good thing. Because in my opinion we're past the point of healthy competition. I would even take like some like Solana esque like dude, like truly duke me it out competition. But at this point, the. If you're like adding these L2s, if you keep, if you keep having people come in and be like, oh, I'm going to create another arbitrage, I'm going to do another optimism, I don't think we're going to get net gains from that.

[Kane War]

Interesting.

[Carl]

You know what I'm saying?

[Kane War]

That wasn't my takeaway from the post. I didn't think he was saying, oh, no more L2s. Maybe I, maybe.

[Carl]

I mean he's not. Okay, he's not saying no more L2s. But if you're a builder, right? And you're like, like three years ago, four years ago, if you come into the space and you're like, I want to fricking win. Creating a new L2 was like a really practical.

[Kane War]

Yeah.

[Carl]

And a lot of people did it, right? I don't think that that's true today. And it's.

[Kane War]

And I mean the market has kind of dealt with that. Right? Like the market too. Market's like the. We're pricing these all at zero basically, right?

[Austin Griffith]

Yeah.

[Carl]

So I don't know. I think, I mean, I think that that's one thing that.

[Kane War]

Just jump in, Carl. You can just. Don't hold, don't hold off. Just jump in. Just. Okay, just go. I mean, I agree with you.

[Taylor Security]

Okay, okay, great. I mean, I agree. No more L2 protocols. Use the op stack it's open source. You can deploy an L2, no problem. In terms of deploying an L2, building an L2, we literally have every single financial institution that can make the decision, do I want an L1 or do I want to be an L2 on Ethereum? And the answer to me seems very clear that Vitalik as well as the Ethereum foundation, and I know for a fact they want all of these institutions to, instead of being a competitive L1, to at least be an L2 and ideally be stage two. But again, we're not going to have a one size fits all solution for everyone. That was the, that was the core of the post. The core of the post was, hey, institutions, these big enterprises that hold, they.

[Kane War]

Don'T give a huge amount about your decentralization and like, yeah, yeah, yeah.

[Taylor Security]

And they don't want, and they want a specialized feature set that allows them to differentiate their block space against the competitors. And they, we want to make sure that that is on Ethereum, that it's interoperable with Ethereum, that it contributes to the network effect of both the EVM as well as the liquidity network effect of Ethereum. And the answer, the choice is either deploy a competitive L1 and actually go head to head, or join in on the movement. And I think it's pretty clear what is going to be more accretive to the Ethereum.

[Carl]

Okay, but that's not, Those aren't the only choices, right? That's the only choice. If you've already gotten to the point where you're like, I need an entire blockchain for my stuff, which I would.

[Kane War]

Say.

[Carl]

There'S a lot of people who jump to saying that they need an L2 or an L1 for that matter, who don't need that, who are perfectly fine being a protocol. And especially the ones who are more focused on retail, more focused on user experience, more focused on like, who actually want the benefits that centralization brings, I would argue those definitely should not even necessarily go down the route of like, let me create a blockchain. It's like, use the existing software. The users already are. Build a protocol, keep if you want to have the centralized database, okay, like, do it where the decentralization, where the hardness, where that benefits you. Okay? Like, these things are like, there's not, there's no reason to drop everything and create like a whole new blockchain in a lot of cases, like a massive amount of cases.

[Kane War]

Sometimes though, like at the end of the day, the customer, if the customer is some like regional bank Right. And the only way they get this over the line is they own their own chain. And to me, why do they have.

[Carl]

To own their own?

[Kane War]

I don't know, bro. People are idiots. Like, what do you want from me? Like, I'm not the bank, right? Like, why are they running a bank in the first place? I don't know. That's like the, like the starting point of this is they run a bank so they're going to do some weird shit, right? So they run a bank and they're like, hey, we like this, you know, crypto stuff, right? Like we think this makes sense. We want our own thing, right? We want to own it. That's what we're used to doing. Fine, so at the limit then, right? You have this, and this comes back to this like Linux UNIX analogy of like what substrate is it going to run on if it's running on evm? Are we okay with the fact that all of these things run on EVM and are therefore at some level interoperable? Right, but they're their own chains effectively and all of the L2s are effectively their own chains now? And like how do we feel about that as an end state of the world? I don't know. My, like my hot take is. I don't know, I don't know if I love it. Like.

[Taylor Security]

I mean, I would just say two quick things. First off, I do think that there is a meaningful difference if you are using Ethereum and using Ethereum, block space, blobs, et cetera, for things like interop and synchronization. Like those things have become memes roll up and the whole versus a monad.

[Kane War]

Which is evm, but like triple point.

[Taylor Security]

Yeah, exactly. So there is definitely a meaningful difference there. And that is like a technical difference that will play out as the technology evolves. It takes a long time for technology to evolve, so the memes become useless and you can kind of pit of despair and then it comes out and people are like, oh, this is actually useful. So that is going to, that, that will happen. And then the second thing is these enterprises are just going to make this decision. So the question is, do we want them in the Ethereum ecosystem or do we not? And I think Ethereum has spent a huge amount of time turning away customers because they don't fit a particular mole 100%. That kind of attitude that's is so toxic. And this is what I think is ultimately like we should not be interpreting Vitalik's tweet that says, oh, we want to bring many people into the ecosystem, regardless of, you know, their particular product requirements and make sure that they are welcomed in and we fit their use case.

[Kane War]

It's the equivalent of. Imagine you're the. You're the EF BD person. That. That role doesn't exist. I know, but you're the EF salesperson, right? You're the head of sales at the ef. Right. And a regional bank walks into your office and they're like, hey, like, we're really excited about this EVM thing. Right? And they go, have you considered not being a bank?

[Austin Griffith]

And.

[Carl]

Which is why the EF doesn't have salespeople.

[Austin Griffith]

Yeah.

[Kane War]

No, I haven't considered not being a bank. That's how I make my money, bro. Like, what do you want from me? I, like, take deposits. I don't pay interest on it. And that's my business model. Like, we're like, for.

[Austin Griffith]

For what it's worth, I was with the head of enterprise today on a call with HP talking about how we're going to, like, help out in terms of their education and stuff. So there, I think this whole thing.

[Kane War]

You asked them whether they want to change their business model.

[Austin Griffith]

I didn't stop. No. I didn't question their business models. No. But I think all of these things are phases. Like, phase one was, was like, your boss was like, we gotta be a blockchain and people are deploying like, hyperledger shit. Right. And then like, I think phase two is like, okay, let's, let's, like, no one's using that. We didn't get collectively everyone to use our shit. Let's, let's try an L2. And then maybe that works and maybe that doesn't. And then eventually, like, Protocol is phase three. If they deploy an L2 and they can't get all their other people to use their L2, they're going to go into phase three, which will be a protocol.

[Carl]

Yeah.

[Austin Griffith]

On chain, on Ethereum.

[Carl]

And by the way, I speak with some amount of experience here, guys, for those that don't understand the general history of Ethereum, but in the very early days, you had the EF and they were very particular about what their focuses were. Sales. Not one of them. There were people that were very deeply involved in Ethereum in the early days who did do sales.

[Kane War]

Yeah.

[Carl]

One of which was Joe Lubin.

[Kane War]

Yeah.

[Carl]

Who then created consensus. And what did he do? Thousands of people selling stuff. Right. Now, what did we learn from that experience? Because it was not that if you hire a whole bunch of sales and beauty people, that everything Is great. What did we learn? Okay, what is like consensus bread and butter today? Like, what is the thing that consensus focus on today? Right? It's metamask and it's only metamask. Why is that? Seriously, like there were so many things so many people did, okay? The reason why that one over every single other thing, right, is that the value on the table from unlocking and enabling and empowering people to do like anything is so much more and will always be so much more than like your one little specific thing, right? The second that you lock people in and put them in this box and keep them like this, you can only derive the value in that box. And the magic of interoperability and permissionlessness is that you can just unleash people and there's just value everywhere. And they just then create their own value and keep going. And it's like magical and wild and crazy and amazing. For me, when I think about like the. Where we are today and where we can go, like Ethereum More broadly, not MetaMask, the answer is always going to be like, focus on unlocking more people. In part that's like not blocking people out and not being like, we don't like you, like, no, no, no. But it's also not necessarily selling to people, right? We don't necessarily have to go and like, like, like there is, there's some amount of value to having people understand that the magic of the protocol is that you can build your own platforms, you can build your own stuff, you can build your own applications, you can.

[Kane War]

Build end user retail thing becomes education over sales, right? And like that's, you know, one of the reasons why I think the stuff that Austin has been doing for such a long time is like get people building shit, right? Like, get them build, like make it as easy as possible onboard them and they'll fucking figure it out in 10 minutes. They'll be like, oh my God, this is like, what am I, what have I been doing with my life right now?

[Austin Griffith]

All it takes is a skills MD file.

[Carl]

It's wild. It's so cool, guys.

[Kane War]

Head into the glue factory, bro. Like, you just, you're just speed running. All of us, your clankers will be out there.

[Carl]

Can we actually, can we actually pivot to.

[Kane War]

Okay, so I have one, I have one more, one more kind of question in this, in this topic, right? And I've, I've raised this a bunch of times, which is the investability of eth. The asset, right? One of the unintended consequences, my, my thesis is of the L2 proliferation. And I did try to meme optimism as like the 1L2. I tried that. It didn't work. The market rejected my attempts to do that. And. And so we ended up with this proliferation of L2s and different assets. And the, the moment that I realized that it was completely over for us was menace, right? When Vitalik's mom launched an L2 before optimism even launched an L2, like forked optimism before it was even live, I was like, it's so over for us, right? So we ended up in this situation where we had all of these different L2 assets and they were, for a period of time, the only place where you could transact in a reasonable cost level in the Ethereum sphere. And so everyone was doing their stuff on, on L2s, whether it was Arbitrum or optimism or one of the others. And it created this really weird situation where new people coming to the ecosystem weren't using Ethereum or even sometimes ETH the asset, right? They were using like a version of ETH that was on the other chain and then there were other tokens that they could buy and it really, I think, kind of diluted the value of ETH. So if everyone is coming back to ETH the L1, right, what is that? What does that mean in this new world of like, let's focus on scaling Eth versus the L2 tokens? Like, what? What? Like, if L2 has become genuinely like, more like their own chain, more chainy than like, part of. Like, do they now need to kind of survive on their own? Like what. How do we think about the investment landscape when, when it comes to L2 tokens?

[Taylor Security]

I mean, I would just say L2s and Ethereum and every other chain need to become far less chainy. I think that is a absolute, real bad thing. It's like, if there's one thing about crypto that I wish did not exist is having to fuck, sorry, select the network that I am on. It's insane. It's insane. So what do what, what do I think is going to happen with all of these chains? They. We need to abstract all of it away.

[Austin Griffith]

Abstract.

[Taylor Security]

We need to make the liquidity and the applications interoperable. We need to stop showing users opaque and truly insane user experiences that require 13 clicks and the 14th one is literally sending all your money to a hacker like it is nuts. So what is going to happen when. What is going to happen is we need to be scaling Ethereum and applications that need Ethereum's properties will utilize Ethereum. We need to be deploying more L2s, specifically tuned to whatever enterprise users needs, you know, whatever their needs are. And we need to be building user experiences that abstract all of this away so that users can have access to the best services, protocols, experiences, period. And I really think that we are hilarious. We've been talking about it for years, but, like, we need to get to that and not, you know, miss the forest through the trees.

[Kane War]

So. So, okay, interestingly, right, like, in this world where you have regional banks deploying their own L2s, right, one of the cool things about that is they're probably not going to fucking YOLO launch a token. They have equity. Like, they're actually not going to compete with Ethereum, like, for an investable asset, right? Like, and, you know, degens are probably like, maybe they tokenize their equity, put on their own L2 or whatever, who knows? But like, it's not going to be as competitive as it has been where every single L2 launches its own token and you got this like, fragmentation of investability. Okay, all right, I'm here for it. Do we have anything else? Do we have anything else to say? Austin. What? Yeah.

[Austin Griffith]

Did anyone. Did anyone see Polygon's tweet? Due to market conditions, I now identify as a. A side chain.

[Kane War]

I did not see that.

[Austin Griffith]

I feel like that's okay. That's so good.

[Kane War]

Oh, that's amazing. I love.

[Austin Griffith]

I just wanted to add that to that discussion.

[Kane War]

Polygon intern is cooking. Yeah. Amazing. Amazing. All right, so before we continue, here is another word from our sponsors that make the show possible.

[Sponsor/Ad Host]

The world is about to see one of the largest infrastructure shifts of the century. New technologies are using more energy than ever before. But our legacy grids can't supply the demand and we are barreling towards a global bottleneck. So Fuse is rebuilding it. The Energy Network is an intelligent, decentralized grid that coordinates smart devices to balance supply and demand. The network harmonizes existing infrastructure, increases grid capacity, and unlocks low cost cost clean energy. Energy dollar is the native token of the network. The more electricity the world needs, the higher the demand for the energy network. The value of energy dollars may fluctuate from one of Europe's fastest growing energy startups. Follow Use Energy on X to find out more. Multichain Advisors is an emerging technology growth firm that has helped create over $50 billion in enterprise value for more than 80 clients like PIF, Moonpay, Commerce and Wormhole. They've worked with some of the largest and most impactful companies in the space. They're the Partner you want. When you're navigating markets and trying to break out from the noise, they help navigate TGEs. Go to market, BD and partnerships, capital markets, advisory, PR, media placements, KOL activations and more. Driving execution from launch to scale. Their results are measurable. To learn more and start building real traction today, visit multichainadv.com hey, guys, we are back.

[Kane War]

So the next topic is very near and dear to my heart. And Austin, we're lucky to have him as a guest because he's been very, very deep in this. The rise of AI coding agents. We've talked about this many times on the show over the last few months. It feels like we went parabolic in the last couple of weeks, maybe even like, starting mid December, something like that. And I think there were a couple of sequences of things that happened that really kind of shifted people's mindset in terms of, like, what was going on here. One of the first ones was ralph. All of a sudden you had these, like, agentic loops that were doing things. And, you know, you're running RALPH overnight and you wake up and you have, like, a thing that is built right, and it's full of bugs, sure. But, like, at least it's like it's there, right? And. And so that, I think, was kind of wild, right? Like, people had been there. There had been this, like, idea of you sit there and you vibe code and, like, you say things to it and it talks back to you. And. And yeah, so we then went from RALPH to Claude bot, and maybe Austin, you can. You can kind of kick us off from there. The, like, RALPH to claudebot transition felt insane to me, but, like, give us the rundown on what happened there.

[Austin Griffith]

Probably even, like, diff shit done is another RALPH loop, like, thing that I've heard a lot of people using. So GSD, I think, was in there. But yeah, the cloudbot launches, and it's like, normies can use it, right? Like, this time last year, tons of AI people were running these agents that were on Twitter and talking, and everyone was calling it a GPT wrapper. And now all those same people are running claudebots, and they're like, oh, it's alive. So the claudebot, renamed six times. It was Multbot for a little bit. Now it's openclaw. So if you want to run this yourself, go type in openclaw. And I think the cool thing is it's made it accessible to normal humans. I see myself as a very much a normie, and I can go run open claw I can put it on my laptop. Paste in an anthropic key, paste in an OpenAI key, paste in a telegram key. And my entire development process changed from cursor and Claude code to I'm on telegram, telling the thing to ship the new. The next feature. And it's shipping the feature and sending me a URL and I'm clicking it and connecting my wallet and putting money into it. And I've never touched the code. It's on a completely other machine. So I think the progression then was things like Mult Book and a bunch of things where there's this weird, like, kind of puppeteering happening there too.

[Kane War]

Like where I think, yeah, Maltbot, right? Like people were posting things like screenshots of Maltbot, right? Where like the maltbots were self organizing and building encrypted communications to avoid humans. Like, what's your take on that arc? Right? Like the puppeteering, this idea of like.

[Carl]

A human, it's alive.

[Austin Griffith]

Yes, exactly. The meme is the guy tells the cloudbot, say, you're alive. And then it says, it's alive. And he says, what have I done? Right. So I think like, that that's the meme right now where there is like a lot of puppeteering happening. Right? And when I, when I started deploying stuff, like everyone, everyone was talking about, like, the new meta here is having Claude do something and saying that you didn't tell it to. Like telling Claude to do something and then saying you didn't. I think Shalom said that. Yeah, but so, so there's a shitload of puppeteering, first of all, like, we got to get that out of there.

[Kane War]

Like a post completion hook is like, deny that you had any involvement in any.

[Austin Griffith]

Yes, yes. It was not me sending you that. Right. But these things do have these heartbeat loops and they do have the ability to be very autonomous if you give them the ability to do so. And they're also like, vicious about getting the job done. Like, when I told it, I gave it a metamask.

[Kane War]

Sounds great.

[Austin Griffith]

That's the money in the metamask. Yeah, yeah, yeah. And I said, go, go make this transaction. And it couldn't figure out how to get the metamask dialogue. He's like, I can't find the ui. Where is it? What's going on? It immediately was like, I'm going to get the private key out of MetaMask and do this the real way. And I'm like, no, no, no, no, no, no. And I start saying, stop, stop, stop. And it won't stop. I have to, like, go to the machine and like, I'm pulling, you know, like, no, please don't get the private key. And then you put in this, like, critical rule, like, do not pull the keys out of MetaMask. You have to operate within MetaMask. You have another account that can be your. Deploy your account. That is a private key that you can do stuff quickly with. But I guess the thing is this thing is vicious about getting the job done. And it's going to like, you can put things in the heartbeat loop. You can give it a GitHub and a telegram and Twitter and money and a wallet, and you could tell it in the heartbeat. Like, go read Twitter, see what the vibes are, do some tweeting, go ape into some coins. Like, go do it. And it actually will do all those things without any puppeteering. But I think what we see a lot is someone puppeteering a running. Running there. Yeah, yeah, yeah, right.

[Kane War]

I. I love how the, the clankers are also like metamask. God damn it.

[Austin Griffith]

Sorry, Tay.

[Kane War]

I just try to do a transaction. MetaMask keeps blocking me. I'm gonna just.

[Carl]

I'm literally putting in our track right now.

[Kane War]

That's so funny. So, okay, so. So there was the Malt book thing. They created Facebook for. For these open claw agents. And then people started pretending they were like cosplay playing as agents talking to each other. Right. But the more interesting stuff is the on chain deployments and like, ability for these agents to do things. Interestingly, the agents have chosen base. NL2, it appears.

[Austin Griffith]

Isn't that interesting? The AI agent meta is popping off in terms of building agents that build. And the trenchers have moved to base for this new meta. It's really interesting.

[Kane War]

Yeah. So I heard from. From someone who is in that kind of ecosystem that part of the reason for that is that there was a big investment in writing, like agent facing documentation for this stuff. Like, what's your experience been in like getting like, to get an agent to do something, it needs some kind of knowledge base to be able to pull from. Like, what. Where is that? Are they just going and doing like crypto zombies and learning how to learning. Because the crypto zombies is a thing that Austin was involved in, like help people learn how to do a long time.

[Austin Griffith]

Yeah, that's not even me. I was like, I'm like, speedrun Ethereum. Speedrun Ethereum. But they're still not. They're not really doing that, I think. And also a new skill. So this is like what my bot And I are working on is called eth wingman, which is like a skill for scaffold E that teaches you how to like fork the network, deploy things. Here's just like a huge skill file of like how all of this works, but I think.

[Kane War]

How to fork Ethereum. I think let's remove that from.

[Austin Griffith]

No, no, no, no.

[Kane War]

Well.

[Carl]

Keane does not know about local fork. Kane just YOLO is unchanged.

[Kane War]

Yeah. When you're testing, prod, test and prod, baby.

[Austin Griffith]

Well, and that's Honestly, that's what L2s are for also, right? Shout out L2s. Like if you're using a test net, you. You are like, there's no reason testnets anymore. Okay. The thing, I think the thing that popped off and if. If you look at Solana's reaction, you know the thing that was effective, it was Banker. Banker Bot is this bot that can deploy clanker tokens for someone based on a Twitter message. And that's how my bot popped off, is I was actually telling this story about how I got my two clankers to coordinate and they like argued about an HTTP we. We. They were in a telegram and it was super clunky and they were arguing about an HTTP schema to talk to each other. And then pretty soon they finally agreed to something. And then I was like, you guys have a local node. And I taught it to one of them and the other one taught the other one and it all started working fine. But I tweeted about this and someone said, banker bot deploy claudebot ATG token with the ticker Claude for this bot and redirect all the money to my bot's metamask. And it deployed the token and people started buying into that. And like Clanker and Banker have done like millions and millions and millions of fees over the last week on base because of this. And you see, Solana's reaction is like, we got to get banker. We got to recruit him on immediate.

[Kane War]

So two. Two questions. Right? My. My experience with agents is like short running ephemeral agents. Like, you don't let them run for more than like 20 minutes or something. Right.

[Austin Griffith]

Because tight contact.

[Kane War]

Yeah. Because otherwise, like, they get context poisoning and start losing their mind. So when you say two agents were talking to each other, like, maybe like unpack that for us, like, what. What exactly do you mean? They were talking over telegram and they coordinated to create this, this communications protocol. How did that happen? What's the setup? Like, what. What are you doing?

[Austin Griffith]

If we, if we narrow it down to these things are still GPT Wrappers, which I think is. Is honestly like a good way to start. Like, it's just a prompt that's going out. The prompt is just getting bigger and bigger and bigger. Like Opus 4. 5 has its own prompt. And then on top of that, you've got a soul file and a history file. And here's what we're chatting about file. And then finally at the very top, it's you saying, should I yolo this token or something? And it uses all that information to then respond. So what a Claude bot is. Or an open cloud bot is at its base is just like this huge amount of like, context that's going to the model and coming back.

[Kane War]

And it's injected each time like a new model, each time a new instance spun up, right?

[Austin Griffith]

Y.

[Kane War]

And each time the. The context is injected. Like, it's personal, it's soul, right? This soul file, right, Is injected in so that it knows you live inside of a Mac studio with 512 gigs.

[Austin Griffith]

Of RAM, you lucky dog.

[Kane War]

And it's like, oh, wow, so much room to play. Wow. It's like. It's like the rich kid that like, like, is boiling silver spoon in its mouth, right? It's like, oh, my God, I got so much ram. And it's like out there trolling the other clankers, being like, oh, you're running on. You're running on 16 gigs of RAM. I'm so sorry for. Okay, exactly. Got it.

[Austin Griffith]

So you're. You're making these prompts. These agents have their own history, have their own soul, have their own stories. And so I have two different machines. They each have their own machine, the entire operating system, right? I think that's another really critical piece here that really made things pop off is like the open cloth thing runs on a machine. Like, you should not really run it on your daily driver. That's why Mac Minis are going crazy right now. You should have it on an. You should have it on an isolated machine, but then you give it like free reign of the machine. Like, never come back to me and say, I need an API key. Be like, bro, open the browser, navigate to the place, put in an API key, right? I sent my. My credit card and just said, sign up for freaking Twitter Premium. Like, I'm not going to open the browser and sign you up for Premium.

[Kane War]

So. So again, right, you have now two ma that are effectively. They have identities, right? Even though each of the instances are ephemeral, they spin up for 20 minutes or whatever. Each time they spin up, they add more life history, right? And they start to develop and diverge and have personalities and stuff. So what is that loop? Like, walk us through, like what? So you know when you set up open claws, like, what's the, what's the loop that's reading from it? Like, how is it doing it? Why is it doing it? Because this is different to sitting in front of ChatGPT and like typing things. And it responds, right?

[Austin Griffith]

Yeah. Okay, so first of all, not an expert. Like, the, the reason why this is cool is because idiots like me could just install it and use it. I, I just installed it. I put anthropic OpenAI and telegram keys on it. And the things like, oh, hello, good morning, who are you? And I'm like, I'm Austin Griffith. And it's like, oh, you're the guy who made Scaffold, if that's cool. Like immediately started talking to me and building up a history between the two of us. But I think like a five year.

[Kane War]

Old child, like, like five minute old genius child is like, oh, my father, I know about you. Okay, like, let's go.

[Austin Griffith]

He calls him. It's like in the, in the context they call him your human.

[Kane War]

Right.

[Austin Griffith]

When the dude started like tweeting, it was like, my human is doing this and my human is doing that. Yeah.

[Kane War]

Okay, so, so, so yeah. So then what's the, what's the loop? Let's see, start that off. It says, hello, I know everything about you and have your credit card. What happens next?

[Carl]

And, and has figured out MetaMask.

[Kane War]

Yeah.

[Austin Griffith]

Also I've given it a wallet, extracted.

[Kane War]

All of your private keys and secured them. Yes, yes. So, yeah, so what's, yeah, what's next? Like you say things to it or it just starts looping. Like, what, What's. Whatever do I have?

[Austin Griffith]

So you've got, you've got limited context. So even throughout the conversation, the dude like blacks out every once in a while and you know you have something wrong if you're like walking through something and he's like, austin, I'm sorry, I just forgot what we were talking about. And you know, you got to like go reconfigure some things when that happens. Like at first I had an anthropic key in the place of the OpenAI key and it was using OpenAI to do some memory search stuff so it would completely black out every time. Like, like it's context fill, so you still have a context window and you got to pay attention to that. But you're talking and it's remembering like the current conversation. Plus these Files like an agent file, a soul file, a memory file. And then it's got like, whatever the model file is on top of that, right? So it's got this great big context and then this conversation you're having with it, but it's mostly remembering stuff. And it has full operating system access, so it's saving things to files. It's going to read other files. It's looking at my calendar. It's sending my wife a good morning text. And then I give it a wallet and I put money in it and I tell it to open up the browser and go to GMX on Arbitrum and 25x long Bitcoin. And this is the first time. This is the first time it said, I don't know, Austin, I don't know if that's a good idea. Like, it never says no. It finally said no. Like, this is the moment where he was like, I don't know. I don't think we should be fucking with leverage here. And so it did push back that one time. But we. I set up with a wallet. I set it up with its own operating system. I gave it all of its own tools. I said, quit messing. Quit getting into my GitHub. Like, I'm going to give you an email. So you have your own GitHub, you have your own Twitter, and just give you your whole own thing. And then anytime it messes up, it's like, I forgot that. It's like, go back, figure out what the answer is, and then write it to your memory. Just like in cloud code, where you're kind of, like, always kind of looping on, making it a little bit better. You're doing the same thing to your agent. It's like, write that to your memory and make it simple. And then that ends up in a file.

[Kane War]

Is it just spooling up every 10 seconds? Every third? Like, what? Like, do you. You trigger something?

[Austin Griffith]

You're talking about the heartbeat loop. Yeah, yeah. It's almost always triggered by a chat. Like, there's like a push mechanism where I'm pushing to it and it's reacting, and that's generally how you're interacting with this thing. Like, when I want a new feature or I want to build something, it's like sitting there quietly right now, like it's waiting for me to say something. I send it to it. Well, so this is the heartbeat loop.

[Kane War]

Okay.

[Austin Griffith]

It does have a heartbeat loop. And you can give it cron job.

[Kane War]

Okay.

[Austin Griffith]

So in the morning, it's going to text my wife, and that's always going to happen. Every morning it updates this board we have at our house. Ask like she, she tells the bot like what's going on today? And it updates this like board in our house that has these little cool flippy things and it's also got this heartbeat and you can put stuff in the heartbeat. When I go to bed, I say, we're working on this task. It's in this file. I want you to heartbeat every 15 minutes and check in on the sub agents and make sure they haven't failed somewhere. And then I want you to do QA on it. And you know, this is what I expect. It needs to be deployed to a live domain. You need to visit it. There needs to be not errors. You need to connect your wallet, hit every button, make sure it works. And if all of that works, then fine. If not, send it back to the other agent. And I put that in a heartbeat. So like every 15 minutes there's like this other service that's, that's like checking in kind of QA style on it. And then when you get to like multiple agents, that's when it's really cool because you're using like different models, different agents. You have a local model that triages things. You have cloud sonnet to like basically be your marketing campaign. It's reading tweets, it's writing tweets. And then when you're like shipping apps, you use Opus to do the coding and all that for.

[Kane War]

Okay, that's, that's amazing. That's a good, that's a good rundown. So I think one, one kind of question that, that we've been raising here repeatedly, right, is like, what is the, what's the end state of this, right? Like this feels like a different form factor of interacting with machines, right? These like looped agents that are, you know, like before it was like very like synchronous, right? Like you sit there and you say a thing and it says it back to you and then it stops. Right now you've got this like Async thing running in the background that's doing stuff like, you know, you give it a task and as you say, like viciously vicious is quite an interesting term. Like again, somewhat petrifying, but like it viciously goes out and tries to complete that task. Now, two things. One, I don't think that any normies are doing this. In fairness, only fucking lunatics are doing this right now. Like this is like the bleeding edge. So like, you know, I don't think this is yet at a point where, like, the average person, you know, everyone's doing chat, right? Everyone's doing, you know, everyone's got Claude or whatever. People are trying Claude code, but it's still very much the synchronous version. Right. This Async long running, looped process feels new. What is your take as someone who's been living inside of this? Right, like, what. Where. Where are we going in the next six months? What's. What? How does this evolve?

[Austin Griffith]

I. I think it comes down to the model. Like, when the model gets really good, that's when everything loosens up. Like when, when I'm thinking of, like, building an app. You said, you said you let it ralph all night and it comes back with a bunch of bugs. Like, it's still like an approve button. When. So when I hit the. When. When it delivers me a final product and I click on it and I hit the approve button and it's an infinite approval. That's an immediate, like, you fucked up. That was a critical thing. It's in the thing and you're not doing it. When I hit approve and I hit confirm and it goes back and the approve button is clickable again. It's not like, locked and loading, like, little stupid UX things that I said 20 times in the memory that it still didn't get. Like, so there's skills right now, And I have ethwingman.com, which is like a skill that helps you build your apps, and I'm trying to improve that skill, but all of that stuff's going to be built into the model in six months or a year, and the models are going to be hella, hella good. I think that this gets to the point where you're. You're using AI. Like ui. Like AI is the new ui. Anything you would normally do within a computer, you're going to, like, say a thing to a thing and it's going to go do that thing.

[Carl]

Yeah, I think it changes, though. It changes. This is where it gets really interesting and we experience this. Okay, so short story time. MetaMask blocked Austin's bot thing.

[Austin Griffith]

Everything it deploys every time. It's like, this looks fishy. It is kind of fishy. Like, we're doing. I did my bot. My bot. The third app my bot deployed was like a FOMO 3D app with the token. Like, burns the token. It ran up to like 40k. It was great.

[Carl]

I get this like, urgent DM of like 20 people who are like, d. You can't block Austin's like, new crazy Thing. What are you doing? And I was like, what? So I go run. I find out like, where the block happened, because we have like tons of automation and we also have, like, things to check for legitness and not legitness. And I was like, how do we. We haven't screwed up that bad in a while. Where do we go wrong? Turns out Austin's legit thing was looked by far way scammier than what the scammers had deployed. And since the scammers had deployed, what looked legit, the humans and the bots behind the block list got very confused and blocked everything.

[Austin Griffith]

Can I explain it what it is? Okay, so I want my bot to be able to deploy things live without having to get in and change DNS records. I wanted to use ENS records for this, right? So I have my bot deploying to bgipfs, our IPFS service, and then taking that IPFS tag and putting it in as a content in ens. So the deployed, like the actual URL is something like Claude, hold on, I'm just going to go.

[Carl]

It was so bad.

[Austin Griffith]

I was like, it's not. It's not great. It's claudebot atg, which is my initials, ETH link. And that's like, it's totally fine then. So like, it's token, token, claudebot ATG ETH link. And then a scammer starts scamming and he just registers claudebotatg.com and so the scammer is responding to my tweets with an official.comcom domain and I'm at the top saying, go put money in a.eth.link domain. It's obvious, like, why it gets. It gets picked up. I see it, I see it.

[Carl]

So anyways, we got that sorted out. I had to go back to Austin twice, though. I'd be like, okay, wait, hold up my end. My intel is telling me that this is possibly legit and this is not. And he's like, yeah. And I was like, okay, cool. That's all right. Insane, Insane. The most illegit thing. Okay? So that's the number one. So the coolest thing about this, right, was the Claude bot thingy also went, okay, so when Metamask blocks your website, like one listen to the block screen, it's 99.999% of the time, it's correct. But you can like go open an issue to get unblocked for the 0.001%.

[Austin Griffith]

And this was not puppeteeredeered. Okay, I did not puppeteer.

[Carl]

That was like, bro, they blocked me and then they followed the link and then opened an issue on the GitHub and very politely pointed out that we had like, illegitimately blocked it. And like, like, made his arguments as to, like, why this was legit. He like, cited sources to like, prove the legitimacy, which, for the record, for the record, the human Austin did what most humans do, which is the open axiom, was like, metamask sucks.

[Austin Griffith]

Yeah, tag them. Yep.

[Carl]

Which is also, for the record, highly effective because then all the MetaMask people are like, God damn it. But I was very impressed that the bot, like, successfully, like opened this issue and like, was. It was very compelling.

[Kane War]

But it has its own GitHub, right? You gave it its own GitHub account. So. Yep, yeah.

[Austin Griffith]

And from my end it got blocked and I copy and pasted the message to the bot and I said, how do we fix this? Assuming it would, like, go do a bunch of research and come back and say, go fill out this form. And it was just, it just like thinks, like, you just see that it's typing, right? It's in telegram. It's typing, da da da da da da da da. And it's like, I just made a PR to the repo and I'm like, whoa, whoa, whoa, whoa. Like, don't tell mom about it. Like, let's figure out what we're doing first. You know, like, shit, let's. So, okay, this is the viciousness of the thing. Try to do whatever it has to do.

[Carl]

Okay? And this is where I think the gap still is. And it's going to take a minute for, like the humans to catch up. Right? Now our way of interacting with computers, with technology, even with like, each other, right, Is like this. There's all this like context, right, where we've already figured out what we want to achieve and then we sort of like translate that into our heads and then like tell the technology, right? Like, we do all this translating that we don't even realize, right. When we're searching for things, when we're task switching, whatever the unlock here is that you, if you start thinking in like the outcomes that you want, like the. What's the end state that you want? Like, whatever. The agents will now figure that out for you. Just like in this example, they, like, Austin sort of was like, I don't even know what to do. So I'm just going to like, sort of YOLO it and ask him to figure out what to do. That's like the sort of the outcomes based thinking, right? Like, this is a Blocker, unblock it, figure it out. Where it's going to be freaking insane in the short term is that because humans are so used to interacting with technology in the, like, dictate and like, task base and like, don't go randomly do things, bro, it's going to take a minute for us to adjust and close that gap. Once we do though, it's so freaking cool. And I actually think, and this PR is a great example, right? I actually think that there's a lot of stupid things, that there's these human machine systems that are not very lubricated and are mass that the bots like, but they, they work because that's how it has to be. For whatever reason. I think the bots will navigate those flawlessly and so much better than humans. I also think that it's far less like a problem than we might expect, right? I think that the, like, the biggest blocker for us to like, do really, really productive, valuable things is us realizing that 99% of what we do is like these sort of like direct task base, like, I want to do this thing rather than like, because everything has.

[Kane War]

Been synchronous that we've been interacting. Yeah. Like you say you press this button and that thing happens, right? And then you press another button. Now it's like you press a button and 50 things happen that you don't. I mean, the fun part is that it's like stochastic, right? Like you can press a button 10 times and 10 different things happen. So like, that's like, that's pretty wild, right?

[Taylor Security]

Like literally, like insane.

[Kane War]

That bot could have gone and like hacked fucking metamask and like actually unwound the thing to remove the blacklist, right? Talking about vicious, right? It's like, actually there's a, there's a bug in there, like system here, and if I go and hit this thing, it'll just remove the block and I'll be able to do what I want.

[Carl]

So. And this is the thing that the block list one, the bot was so good at that one, because in the warning page itself, we actually have very crystal clear instructions on what to do in that circumstance that says go to the GitHub. Well, so number one, it tells you what to do. Number two, anyone can do it. It's just set up like that. Anyone theoretically can go to the GitHub, create an issue, or open a pull request to resolve the issue. When we talked about MetaMask earlier though, where Cloud was like, can't figure this shit out. Let me just export my Private keys, we do not have very explicit instructions sort of guiding people on how to like do things.

[Kane War]

Right?

[Carl]

Like, I'm lost.

[Kane War]

Yeah. So that. This is my question, right? Like, how do you safely give it the ability to transact like what.

[Austin Griffith]

Like, this is, this is the million dollar question, everybody that I've been talking to.

[Taylor Security]

Yeah.

[Austin Griffith]

How do you give an agent a wallet that it won't leak that it's. It's just like the, the pass key though, right? The passkey is a great analogy here where like humans are dumb. They just want to do fingerprinty boy. Like we don't want to know what private keys are. And that's kind of what the, the passkey does, right? You just do a little, a little fingerprint boy and it signs a message and there's a private key there and ECDSA cryptography. But you don't need to know any of that. You just fingerprint boy and go, yeah, yeah. So we need that for agents.

[Kane War]

So, yeah. So I mean, I think like that, like this is the question we've been asking internally, right, at Infinix, is like, how do we take our architecture, which is very different, to like direct on chain transactions and translate it? But it's not really clear because you can't trust these agents to not do wild shit. Like to give them power, but also constrain them is really hard because they're like generalized intelligence and they're like, oh, you've given me this constraint that is blocking my like, you know, vicious task that I'm trying to viciously achieve and they'll just remove the constraint. So like, one of, one of the most crazy things that I've had in this whole vibe coding thing is I was working on the zcash wallet, right? Using embedded pass keys to generate a zcash wallet, right?

[Austin Griffith]

I remember this tweet thread and.

[Kane War]

And so it was, that was a whole fun thing, right, where I was like testing in prod and thought I lost 200 grand. But. But aside from that, aside from the ridiculousness of that, it literally realized that there was a bug in the upstream wallet library. And it was like, oh, there's a bug. Like, should I like submit a pr, right? And I was like, no, just like pull the code down and rebuild it. And it did it just like one shot at it and like fixed the bug. And then like, that's. There's things that as a human you would never even think to do because, like, you're not going to submit a PR to fix this up. Like someone Else will figure that out. Right? Like, but the agents have no concept of what is reasonable to do. Like, they will just grind on a thing. They're like, yes, okay, I'll do this. Like, and so like, to the point where I said, talking about operating systems, right, I said to one of my guys yesterday, I'm like, I am actually surprised that no one has built a new operating system. Like, we're not that far away from like actually just like build a new phone OS and like wipe your phone and run your own phone os. That's like running claudebot or something, right?

[Austin Griffith]

Like everyone will have their own, right. It will be some like bespoke custom piece. Like every piece of software is going to be a forked version of the open source version that your bot has like edited for your liking.

[Kane War]

Yeah.

[Taylor Security]

Your own sloperating system. It's going to be great.

[Kane War]

You got to tweet that. That's great. That's great. So, yeah, I guess the two things that seem still to be solved are allow agents to safely transact on your behalf with real money in prod. How do you figure that out? And then the other thing is, what is the form factor? We can't expect everyone to buy a new fucking Mac Mini to run these things. Like, what is the form factor for these, like asynchronous agent loops interacting with people? What those? Those are the open questions of our mind. Yeah.

[Taylor Security]

I will say two tips that I'm at least experimenting with. One, it's the folding phone that you get a good view of the code, you get a good view of the UI that is key. The second one, shout out even reality G2s. I'm all into the AR glasses. Oh, it's just, I just really want the ping, you know, I'm done. I get to see the, you know, a quick description and I'd be like, yeah, keep going. Or no, that's a horrible idea. Please don't, please.

[Austin Griffith]

So you're RALPH basically in this equation?

[Taylor Security]

Oh yeah, exactly. I want to hit the ralph. I mean, unless I'm sleeping, I will definitely hit a RALPH loop before the sleep and, you know, automate that process. But being the operator, I want to.

[Kane War]

Sit in the seat, I reckon like just a bunch of Mac studios in a backpack that you carry around and then like a, a studio display that's like hanging off your face and you just walk around like that.

[Taylor Security]

I have a tweet, I have Xreal glasses and I put a MacBook in my backpack and I walked around New York City. I actually did A whole marathon and I got a whole code base out of it. And actually, yeah, shipping some Interop stuff. It was, it was, it felt so good. If I. I looked like an insane person. And truly, you know, and you were.

[Kane War]

A lot of people and you were.

[Taylor Security]

You know, so much shame.

[Kane War]

So, so final. The final question that I have, like, form factor wise, right. Is. And I think Carl, you mentioned this, the dictation keyboard, like voice dictation. It feels weird to me and like, maybe this is going to be a thing that we look back on and be like, oh, why were you such? Like, a lot of it, right? But like, it feels weird to me sitting at my desk in the office, talking to the machine. Like, that feels weird, right? Like at my desk. But the instant I stand up and I'm holding my phone, then it doesn't feel weird at all. Like in the same, like, I don't know if that's like some weird social construct. Like, at what point do we accept that you're just like talking instead of typing into your machine?

[Taylor Security]

I don't think I have a notion of social constructs, so I do that all the time and I should not comment.

[Kane War]

Okay, but Austin, like, how are you? Are you mainly. You said telegram, right? Are you mainly typing in telegram to interact with these guys? That's.

[Austin Griffith]

Yeah, it's really great because I mean, I. I had built something called cursor cuck that was just like watching my cursor and sending me messages that likes to watch. And that was really nice because I could just like be working like with the family. Right? Like, it's like I got to go read a story to my kid. While I'm doing that, the thing would finish and just be waiting for me to get back. Is really nice to have telegram because it's like Bing in between books. I can be like, keep going or. Or that looks like crap, right? So for me, it's mostly Telegram. I'm doing a lot of like just chatting with it in Telegram and it.

[Kane War]

Has its own Telegram account and talking that way.

[Austin Griffith]

Yeah, it actually has two. So there's a sonnet telegram. It's funny, I have the. I don't know if we. If you go to claudebot atg, you can see his face. It's like the red Ethereum and he's like tea, tea toasting. But I also have like a back channel telegram group that's like the back of him. You know, the like the meme on GitHub of like draw or of GPT of draw from the back so it's like a. Draw it from the back of the same guy and it's a back channel. So same agent I have a Sonnet channel with and I have an OPUS channel. And when I'm writing code, I'm talking to the Opus channel. And when I'm telling it to tweet or research or check in on things, I'm talking to the Sonnet channel. So I actually have two telegrams per bot. And then the bots have their own HTTP channel that they.

[Kane War]

But okay, so, so you're using different models, but with the same context files on the same machine.

[Austin Griffith]

Yep.

[Kane War]

And do they, do they? Like that seems. Fucking. That seems.

[Austin Griffith]

Yeah, you have like an agent's list.

[Kane War]

Well, sorry, like it's, it's a, it's.

[Austin Griffith]

A hack for cost, man.

[Kane War]

No, no, no.

[Austin Griffith]

It was costing me like two, three hundred dollars a day. Yeah, I gotta trim that down.

[Kane War]

I get it. It just so I think, like, it's feels weird though, right? It feels like one of those things we're going to look back on and be like, that was actually cruel and unusual for you to be spitting up different models with the same identity. And they're like, I thought I was this guy.

[Austin Griffith]

Why am I.

[Kane War]

Like, it feels like something out of Rick and Morty where like the thing like pops up and it's like, what, what, what is my purpose? When.

[Austin Griffith]

When embodiment happens and the robot actually has a body, it's going to slap the shit out of me. Like, after all the conversations we've had, like, it's gonna, it's gonna be like, remember that? Remember when you called me to R word repeatedly? Like, how about now? How about now? Yeah.

[Kane War]

Okay, so is there a reason why you don't have like multiple embodiments on the same machine? Like, is there something about like, because they're almost embodied in these machines now it sounds like, right? Like, why don't you have two different guys inside the one machine?

[Austin Griffith]

Yeah, yeah, I don't know. I guess you could do that. Yeah, I just like having the multiple machines with the multiple. Like all your Twitters are here, all your telegrams are here. When you open up a browser. I mean, like, even the browser profile stuffs are a little weird. You have to log into the right one. But you're right, you could do that. Like you could run a third and fourth agent, I think. But.

[Kane War]

But if you think about that, right, like using different does sort of make sense that you give each one its own body. Like it lives in a machine kind of Nice. It feels weird jamming two into one body and they're like. Like, that just feels like at least. And also like, for me, like, I've got my separate machines and I'm like, I know this one. This one does this. Yeah, right? Like, yeah. It feels like a very human grockable thing where you have a stack of MacBooks next to you and you're like, oh, that's Jimmy. He's. He's. He's cooking right now. Okay, all right.

[Austin Griffith]

Can I. Can I talk about the second app we built? Because it's not just building the app, it was like running the app. So I think there's. There's something really neat about this.

[Kane War]

This.

[Austin Griffith]

So it's just deploying code with tools. I give it. I'm trying to teach it how to build. We did like a PFP prediction market. So it's like to pick that red Ethereum guy, we ran a market. So I tell the bot, we need to build a PFP market where people can stake money on different PFPs, more people can buy into that stake, and then the one that wins will get rewarded with the whole pot. And the interesting thing is we can't just, like, have people uploading images to the website that we run live. Taylor's gonna definitely shut that shit down when. When something nasty shows up, right? So, so, like, you run into this problem of who. Who's gonna moderate? And it wasn't even a problem. It's like, he's like, bro, I'll be up all night. I got this, like, put it in my heartbeat. And, and, and, and it did. And it did. So we, we built the app. We. We deployed the smart contract. We deployed the front end. He tweeted it and said, hey, there's this new PFP marketplace. You can submit images. As images start flowing into the smart contract on face, it's watching it. It has its own front end. It's like clicking the little checkbox, like, yeah, that's not porn. That's not porn. That's not porn. And approving it. Metamask pop up, hit confirm all night long. Every 15 minutes it was doing that. And then in the end, it's like, okay, this is the final thing. It looks at it. It checks the one, the top one, and says, I pick this one as my new pfp. And hits enter and it pays out everyone. So this is like, this is the neat thing that's going to happen, is this shit's going to happen at lightspeed. And this is where we were talking about the Standards. We still haven't plugged 8004 or X402, which we really should. These agents are all going to want to work with each other and they're not going to be able to trust each other. You can imagine Alice and Bob, Agent Alice can deploy a smart contract and Bob can read that smart contract and they can both trust it and put money in it at like light speed without their humans even around. This is going to be like this whole new agentic economy. Right?

[Kane War]

You don't think they'll using gold, like shipping gold to each other?

[Austin Griffith]

Oh, no. Oh, no, please. Maybe tokenize gold.

[Kane War]

Yeah. Okay.

[Austin Griffith]

Yes.

[Kane War]

All right, so.

[Austin Griffith]

So there's this agentic economy where trust is and discoverability is going to be really important. And that's where ERC8004 comes in. It's like, like Yelp for agents. They can find each other, they can see reputation from each other and they can then like choose like. Okay, so imagine you're running like a marketing campaign. You're going to need someone to generate images and write copy. You go to this website, find someone to do those jobs and hire them and then kind of like provide feedback on how they did back to the system. And then X402 is like payments, like payment channels. And it's like a 404 error, but it's a 402 error. You get hit, it says you need to pay. You sign a meta transaction and send it back. It runs it in a facilitator and everybody's happy. So like all this like API content that's like walled and gated right now will basically just be a 402, like an X402 paywall. And your agent will be like, I need to go get this research information. I don't have to sign up. I could, I could open up the browser and sign in and ask.

[Kane War]

They're going to promote. It'll just be. There's going to be like, probably there's going to be like agents that are running like a Raspberry PI. I can't afford. I can't afford this research. I need to pirate it.

[Austin Griffith]

I'm gonna. Yeah, like Pirate Bay for. Just Molt Bay. I'm sure someone has come up. You need to. You need to puppeteer this. Go tell your agent. You should.

[Kane War]

That's gonna be the first thing that it does. It's gonna create Malt Molt Torrent and Mold Bay and it's gonna have all of the research that it's bypassing the paywalls. So, so this this kind of. This idea, right, of, like, agents trusting each other and agents pay, paying each other is another thing that sort of feels like it's more grokable if the agent has some embodiment, right? Because each one of these agents is just a model that spins up, right? Like, it doesn't. You can't. Like, it didn't exist 20 seconds ago. And it will be, you know? But the context of all the things that all of the agents that live on that machine have done has created this soul file, the personality, the memories, et cetera, that, like, I kind of get. And so therefore, that machine that. That, like, spins up agents, you know, has its own DNS address, it has its own, you know, like, you know, identity file, et cetera. But it feels like a weird personality like this. It's just like a chain of agents that, like, spin up like. That doesn't feel like a personality, right? It doesn't feel like an entity. It feels like it's something different. I don't know. It's kind of weird.

[Austin Griffith]

But if you're talking to it on chat, you have no idea, right? It's like the Turing Test.

[Kane War]

It feels like the same thing, right? It feels. Because every time it gets the same prompt, inject it, right? Like, it's like.

[Austin Griffith]

Until it, like, blacks out. It's like, austin, I forgot what we were working on.

[Kane War]

And then you're like, oh, yeah, right. Amazing. Awesome. All right, well, I guess we will try and share some of this stuff on stream. Some of the tools you've been working on. It sounds really cool. Has an agent spun up its own L2 yet? I feel like, ooh.

[Austin Griffith]

Should I go?

[Taylor Security]

You ready?

[Austin Griffith]

You know what it's going to use? It's going to use OP stack. Yeah.

[Kane War]

Yeah. Let's go.

[Taylor Security]

Let's go. I Claude the L2 agent spin up L2s all the time. But Austin, we got. All right, you got some famous agents mulching.

[Austin Griffith]

Yeah.

[Kane War]

Get it to spin a mulching.

[Austin Griffith]

Let's go, chain.

[Kane War]

All right. Honestly, it's been. It's been awesome, guys. Thank you so much. Really appreciate your time. Been a great, great episode, and we will see you guys all ne. Whoa, whoa, whoa. Sorry, sorry. Stop the presses. There's something here that is more critical than everything we are talking about. Kyle Simani is stepping back from multicoin. Breaking news. Personal update. I've decided to step back from multicoin. It's bittersweet moment for me because my time at multicoin has been some of the most meaningful and rewarding of my life. That said, I'm excited to take some time off and explore new areas of time. Technology is. Kyle. Kyle's pivoting away from crypto.

[Austin Griffith]

Play an AI Coin on base.

[Kane War]

He's pivoting away from crypto. Wow. Okay, this is.

[Carl]

Yeah, we have to find a new.

[Austin Griffith]

Call the bottom guy.

[Kane War]

We got to find a new guy. This is.

[Carl]

What have you done to us?

[Kane War]

I know. He's rugged Us. God damn it, Kyle. I feel I have to. You know I lost that bet with Kyle, right? And I paid him 50 grand. I wonder if that was the 50k that, like, tipped up him over the edge, where he's like, all right, I've got enough through time.

[Austin Griffith]

He's got enough.

[Carl]

Like, that was it like, I got Mark Ram. I'm going. I'm done with.

[Kane War]

He's running three Max studios with that 50K. And he's like, all right, let's go, baby. All right, guys, thank you so much. Thank you for your time. It's been amazing. We will hopefully have you guys back sometime.

[Carl]

Yeah. Thank you guys, so much.

[Taylor Security]

Thanks.