Shaw: The Rise of ai16z’s Eliza—Crypto × AI Agents, a $2.5B Valuation, Trending Globally on GitHub

Transcript

You're now plugged in to the Delphi podcast. Everyone, welcome back to the podcast. I'm Tommy, your host, and today I'm thrilled to be joined by Shawn, who is the founder of Eliza Labs and does a zillion events. I don't know how he keeps up. And I'm also joined by Ejas, who's a friend who just started his fund, 26 Crypto Capital, and he's been intimately involved with the liquid crypto AI scene. Uh Shawn, how's it going, man? Good to Good to have you on. Yeah, man. Thanks for having me. You know, out in Seoul, having a lot of fun, on a little tour. So, I'm glad to connect with you again and have a chat.

I love it. Ejas, you gave Shawn a different title than the one I introduced. Could you repeat that? That was pretty good. Yeah. Before we went on There's two titles. The man, the myth, the legend, and Degen Spartan AI's daddy. Yeah. I'll take that one. It's such a good phrase. You're You're a dad to a robot. I've always wanted to have AI children. So, uh so Shawn, it's great to have you back on. Um you uh you shared a pretty crazy tweet I want to start off the podcast with was you kind of shared your your journey from, you know, being you, being in debt, living in a small apartment, you

know, wife getting mad at you, selling this thing out, getting some tokens. Like, could you maybe just walk us through the quick rundown of of how you got here? Cuz the story is pretty crazy. Um So, like, I I've been an open-source dev for a long time. I I'll start with one really crazy thing I didn't even mention there, which is um I I made an open-source project that I really believed in. Uh Open Spatial Web project, like end-to-end ability to launch worlds and put characters in them and do that whole thing. And I launched it with somebody who deleted me from the GitHub and sold it for $75 million, dollars, I got nothing.

And I was the chief dev, and he did not code a single [ __ ] line of code, but you know, sometimes that's how it works. Um and oh yeah, I'm definitely suing him, but it's been a long time grinding. And so I kind of went from that to like, you know, um being in a pretty bad situation of uh kind of losing everything. Uh and having like uh my entire reputation totally trashed and by somebody who made up a lot of [ __ ] And I had to go from that to okay, well, what do I do next? Um and it really hurt, you know, it really hurt to let like I think a lot of you asked me about like, don't you worry about your source code being stolen? And I'm like, you have no idea. You just have absolutely no idea.

Um and so from there um I kind of picked up the pieces and this was like kind of right at the beginning of COVID, and I just was like, well, what's next? And I started working really hard on um the AI agent stuff. I kind of been working on already, but I kind of had to start over. And uh uh you know, it was it was really hard, and I just um I I had um I'd taken some money as an investment from a friend, and I worked really, really hard to try to meet like meet the sort of milestones of that and make it happen. Um and we we kind of did, but also like drove myself pretty badly into debt doing so just because um this person had taken all of the money, and I

took all of the responsibility to try to ship what we had promised. Um and was like basically doing a lot of like service projects, helping people, clients, kind of stuff while trying to make this thing happen. Um and at some point it was just like, this is, you know, the kind of the bear market hit, and Metaverse wasn't so hot anymore, and it was just like kind of like not the right thing. Um and uh you know, that was a hard time. So I uh I ended up kind of going to work at a as lead dev of a company called Weather Verse, and I um that was pretty good. Like those guys were good to me.

Um but uh project got hacked, lost the whole treasury. Had to pivot. Uh it was pretty pretty dark days. You know. Um and uh That's like crying in the shower type moment, for sure. there's been a lot of setbacks, let me tell you. Um and I I I think a lot of uh yeah. It was a lot. And um uh kept building. Um got, you know, started a platform with the creator of Project 99 called Magic. And we we raised a little bit of money. Um um about, you know, like kind of an okay seed round. And kept building that, kept building that. I sort of kind of felt like we really needed to focus on like

an end-to-end agent. He really wanted to build a platform, like a no-code platform for people to come and build agents on. And I I just felt like if we gave them the whole thing, they'd copy it. But if we didn't, they wouldn't really know where to start. So um uh we kind of got close to running out of money. And I was like, I just I just want to go work on the agent stuff. And I'd actually already started building the first version of Eliza on that platform. And um my actually my wife said This is insane, Sean. It sounds like you've just been kind of trying things, beaten down, trying things again, beaten down. And this has been like kind of like a consistent part of your persona.

In my perspective, I kind of like tracked what you've kind of put out on Twitter. And you've just been relentless, which to be honest is like a hallmark of of any kind of like top founder in in any kind of space, right? That's insane. Um you mentioned earlier I and I I I want this to kind of like pivot into this to kind of like demonstrate how you are this consistently. What Where are you right now? You mentioned you're in Seoul, right? What are you doing in Seoul? What have you been doing? It seems like you've been in Asia for like the last couple weeks. And it's been non-stop like meeting with developer communities out there. Like can you tell us a bit more about you know, what's been going on? Like you're you're now at the center of this AI agent matter, right? This

A16Z stuff. Um and you're still on the ground like kind of like getting [ __ ] done. Like like what's the what's the move right now? Like why are you in Asia? Um when we launched this project, yeah, I've been throwing a lot of darts against the wall and just like I I think anybody you talk to who knew me from then is like, "Yeah, Sha is like the unluckiest and hardworking the most hardworking person I know." I I wouldn't say I'm the smartest or anything else, but I'll definitely take the the I I worked really freaking hard and didn't really try to grind this thing out. Um and once we kind of this A16Z took off, um I I got a lot of like a lot of messages from people over in in Asia, um especially China, but generally all

over. Um some of that was like a lot of fun. Uh but but generally, um it seemed like we had a lot of supporters over there and I I did um I met some people who are part of this community called the 706 community. And one of them came in like helped us to moderate our um our like Chinese channel in our Discord. And we did like a little hackathon together and I got to meet all these devs and judge their projects. I was just like, "I got to go over there and meet these people." Um and so we just booked this little uh like, "Hey, I'm going to these cities. Uh I want to meet all the devs. I want to meet everyone who's interested in our project." And people just like came out of the woodwork and put these events together. Um and I've just gone from

event to event like talking, meeting people, just just you know, it's crazy. Like crowds of people just getting, you know, sitting there and meeting every single person and talking to them about their project, sharing my Telegram and saying like, "Let's let's do something. Let's get in touch. Like just give me a little bit to catch up because there's a lot." Um and uh and now I'm right now I'm in Seoul. I was in Hong Kong and before that Beijing and Shanghai. Going to go to uh Japan after that. Have you seen anything crazy from some of these meetups yet, Sha? Like has someone come and pitched you like maybe like a wild idea or maybe they pitched you something where you've been like, "Huh, I didn't even think about that.

That's super cool that you're building that." Oh, yeah. I mean, I've seen a lot of interesting games. I've seen a lot of interesting like applications. Obviously, a lot of like waifu stuff, virtual girlfriend stuff. It's kind of interesting. I was talking to a guy today here in Korea. He was like doing doing robots and I I you know, several people doing robots, which I think is interesting. People doing wearables. Um a lot of people talking about, you know, how well we have these like data collection platforms and we do these like like fine-tuning and and like data labeling and stuff, which is all like could really tie into like we have the swarm and we could, you know, do some really interesting things with this very

unique data that the swarm is collecting. Um I I think it's all like still pretty early days, but definitely like incredible amount of enthusiasm. People doing kind of every project. I think that the the the boring stuff is what I'm probably most excited about, which is DeFi. Um a lot of people building AI agents into their DeFi protocols as like a way to reach users and explain and lower the barrier to entry. I think that's I I said this before. I think that's really the killer app of like the next the next meta in the next few months. Um and yeah, I mean, a lot of yeah. Yeah.

You should know, Shawn, this is a great a great thread to hit on, right? Like AI 16Z is worth several billion dollars now. There's thousands of agents like deployed using the framework. If you go to Eliza's world, it's like a really easy way to see, you know, what you have built and and what's being built. And the developer interest is just off the charts. Like you guys have been trending globally on GitHub, which is nuts for for weeks if I if I remember correctly. And now it seems like we're sort of getting into this narrative where, you know, people are tired with these influencer auto reply agents on Twitter and they want agents that are actually doing things, right? Like creating tokens, managing economies, managing ecosystems, like doing DeFi actions.

Like is this where you see it going? Like is it is DeFi where you want Eliza agents to focus on? Or or yeah, what would be your view there? man, I'm tired of the reply bots. I I think that there's such an obvious business opportunity here and it's just a lot of people who are, you know, like just download the thing and shill it pump the token and and I really want to see us get beyond that. Um I think the things that I'm most excited about fall into kind of three categories, which are agents that make you money, obviously, uh agents that uh bring their product to the right people, like find the customers that are looking for those products and bring them to them. Uh and then agents that save you time. So, I think right now like we're stuck in this

like slop bot reply meta. Like I I just block all the bots that reply when when they're not summoned. And I think that I I'd actually encourage people to do that cuz I think it'll create like a social back pressure that forces agent devs to actually think a little harder and and build something real. Um if you're just like following KOLs and and commenting on everything that they're doing, you're actually making it worse for everybody. Um and it's not going to help your token at all. Not any more than like the random, you know, 200 follower PFP guy who's like, "What about this token? What about this token?" You know, like come on. Um and so um what I I I definitely uh I think I'm most excited by DeFi because it's like any anywhere like

there's a lot of arbitrage and anything that's like very valuable and like very obviously could make people money, but they just don't know how to use it. And I think that DeFi really falls into that category more than anything else I can think of. Um a lot I I I kind of realized this in Yeah, sorry. Go for it. No, what do you think some of those things might look like, Shawn? Um I let's start with like liquidity pools. Um like a lot of people have a fear of using liquidity pools at all because of the impermanent loss that can be associated with um the tokens, like, you know, your pairing kind of um going out of range um or having a very wide range and then um you you people just don't like really understand how liquidity pools work. And so, uh we've been working with people

like uh Orca, and I'm also working with another team uh who's building something on Meteora with the DLMMs. Um and the bot a- automatically kind of looks and sees the potential of, like, okay, cool, I could um these these tokens are going out of range, I'll pull them out, I'll rearrange them. Uh can also, like, send the yield back to your wallet, and just kind of like automate the entire experience in a way that's relatively safe for users to um put their tokens into. I think part of this is also the meta of meme coins is like so volatile. Like, actually like at the beginning of a meme coin launch, they pump so hard that it's kind of like hard to LP them, but once they sort of gotten stable, the volatility like really plays to your favor. Um and there is a lot of money to be made.

And frankly, like, I don't sell tokens. I I pretty much just make money off of LP. Um and I've tried to encourage a lot of other agent devs to do this, but I've been shocked by how many like you know, I I had somebody a friend come to me and he's like, "I'm really struggling. I'm not making any money." And I'm like, "Well, have you like liquidity pooled your tokens or done anything?" He's like, "No, I don't have time." Like, dude, what are you you're doing this wrong. Um you have like a whole supply of tokens. You should be doing something. You could be earning a lot of money off of the volume of your token. Like, you have no liquidity, and then you could be liquidity.

It's almost like a like a yield strategy, right? Like, if you were to like kind of dumb it down, if you could pitch it as kind of like an automated risk-managed um yield strategy, then I think a lot of people would kind of be more prone to using this. I mean, sure, you've said this over a number of different tweets, and I completely agree with you. The whole point around these agents is they're going to make these things these blockchain pieces much easier to understand and use. They're going to abstract away all the complexities, right? And the LP pool within DeFi is one of those things. I agree with you.

These agents are almost kind of like make these markets more hyper-efficient, right? Um what are you seeing aside from just LP pools? Like one idea which is you know, kind of like a popular idea right now is that um you know, these agents are just going to or like manage their own funds. They're going to ape, they're going to degen, they're going to trade. And obviously, you know, AI 16Z, you've got uh PM AI Arca, you've got Degen Spartan AI which are going to trade their own kind of AUMs. Like what what where do you see kind of that developing? Do you think we'll see that in, you know, Q1 or this year or all? Do you think these agents are smart enough or capable enough or what do you think?

Oh, we're definitely going to see a lot of it. And um I think that LLMs probably aren't the best thing to just do trading on right now. I do think that if you had the right like APIs to pull market intelligence from, um you could make reasonably good assessments. I mean, if you look at like AI XP TI, I read something like he's like trading at like 41% success on his calls or something like that, which is actually probably pretty good considering like most coins don't, you know, don't necessarily do well. Um but I think that I I think like uh LLMs are are not really very good at like complex decision-making. And the more you context you provide to them, the more real reasonable they are, but they are just like predicting the next token. I think where they become really valuable

is in turning unstructured data into structured data, which is like turning a group chat of people shilling to each other into information that you could then um do something with. So, like our our core um research, like we we have a team of researchers, like PhD researchers, working on a paper. And that paper is called a marketplace of trust. And the idea of that is, well, if we took all these people in that chat or or like all the KOLs on Twitter or whatever, and we actually took that face value and made trades based on uh their like based on the recommendations, then would we have made money? And for some of those people like some small percentage of those people like are definitely better traders and recommenders than the other people. And we're looking at who are the

top percentile of those people and let's just listen to them, you know, cuz and so this is kind of like prediction markets like there's just some group of people who are super forecasters who are just crazy good at prediction markets. And then the vast majority of people are pretty bad or they're just like really, you know, um sort of subject to like pretty basic behavioral economics and stuff. So so that that idea is like how do we actually get agents to trust people? Well, it sort of has to like track them by metrics that make sense for whatever it's doing and then leverage that as kind of its trading strategy. So it's it's like definitely more of a collective intelligence play than like a oh yeah, let's just like give the a bunch of listings and time series data to an LLM and have it say oh

yeah, I'll buy this or that. I think that there I've seen some people doing that. Um we certainly could do that, but I just don't feel like it's really what the technology is great for and I don't think it's really solving the hard problem of of that we're, you know, of of like how does an agent know how to trust somebody or how to trust people in a community and stuff. And I'm really interested in not just applying this to money, but to apply it to governance, um rewards for contributorship, and these other kind of more abstract things, but money is definitely the easiest cuz it's kind of like a nice Lego block. It's very easy to measure. Um and I think in and so we want to kind of do something that's a little bit more flexible and and reliable than just like cool, yeah, throw this stuff in an LLM.

I definitely think you could do like it'd be really easy to build that agent, right? It's just going to buy and sell stuff. I just don't know if it would actually make money. I assume that it would if it was buying like top 10 tokens in a in a bull market, but I don't think it would necessarily just by taking like any shitcoin off pump.fun and like trying to predict if this would go well or not. If that makes sense. So No, it's a great It's a great take. I've always wondered if people would keep that that bot internal. Like, would Ren Cap ever open source their their like trading algorithm? I I guess I'm curious. Like, if there were an agent that was really good at trading, what is the incentive to like open source that, create a token around it, versus just trade for yourself, right?

Like, what are What are the community benefits, or or are there not any? I I totally Like, so there's a company that I won't name, but they sent me something like, "Yeah, they can predict token price like at 70% accuracy." And I'm like, "If I could do that, I would not be here. I would not be telling you about that. I would just be printing infinite money." Because, you know, 70% accuracy on on something that's like shortable, like Bitcoin, like you could definitely um you know, you could make infinite money. And I'm sure that like that's what the BlackRock's of the world are doing to some degree. They're they're trying to process all the world's data to to make these insights on stocks and stuff. And maybe they are successful at that. Um they certainly have a lot of people working on, you know, that kind of

thing. Um but I don't think in this world of like very low market cap stuff, which is really like subject to behavior, and just like, you know, any like Ansem retweeting the thing is probably going to have more impact than like any fundamentals you could have predicted. Um and so it's probably comes down to like, you know, somebody posting like, "Hey, here's a contract address. Ansem just retweeted it." And if you can respond to that immediately, that's like almost better than any algorithm you could really invent, I think. Um And that's actually why I think meme coins are interesting as you know, Yeah.

Yeah, you know, it's like a super low market cap thing that's super sensitive to social dynamics. So, if you can just follow the social dynamics, like you can get That's kind of where the alpha lies. Yeah. Um I'm totally with you. I I guess I have one longer question for you. So, my friend Vishal from Mythos VC, who we we invested in, had a thesis pretty early on that the only way you really differentiate an AI is having, you know, proprietary data to fine-tune a model, um a totally new model architecture, or like a different or unique business integration like Microsoft Excel with with their their bot etc. So I'm curious like looking at the Eliza landscape and looking at all the agents you see and all the developers working what do you view as the major differentiator with a team using Eliza to bring in insanely

new novel agent to market like is it in the model is it in the data is it in what Eliza provides like what is it? There's definitely a meme going around of like you know isn't it just a GPT wrapper? It's like well I mean like is a website an HTTP wrapper? Like is an application just like a you know I don't know a react wrapper like like it's actually more about the product and more about like do you have customers using this product giving you money? Like that is the fundamentals of anything. I think models are extremely commodified they're extremely expensive a base model from scratch like in the hundreds of millions of dollars

and and so yeah I'm sure if we had like 150 billion dollars like open AI's market cap like it would be pretty trivial to say like okay cool let's build the whole end-to-end training and train models but then like what's the maximum benefit of that? Okay now we're competing with meta open AI XAI Google on who are all just like trying to edge out that extra like point two percent on the MMLU benchmark to prove that they're like the best model in the world and and at the same time there's there like XAI is open sourcing the last version of Grok every time they release the new one meta releases everything they make and they're just open sourcing and kind of doing it like a mind share dominance and and they really have like taken over the

mind share of LLMs across the board. Open AI's like basically chat GPT is an open AI wrapper like I could build chat GPT and sell it. The main thing they have going for them there is like brand. Um and they do have, you know, obviously great products and stuff. And I just think that it's like the wrong space for any of us to compete in. I think we should be focusing on enabling developers to build products. I really think that this is about the internet. This is about the future of of like websites and products and how users like use applications. Um and it it really like there's so many great like this is kind of the same thing like why DeFi is killer of agents, which is there's all these great products that already exist. The infrastructure is already developed.

They're just waiting for users and they don't know how to find them. Like you cannot Google DeFi protocol to make me money. Maybe you can find some list somewhere and you can do your research, but like it's surprisingly hard to to be aware of this stuff if you don't know what to look for. Um And so I think that where the real value proposition is going to be connecting a lot of stuff that already exists and just changing the paradigm and moving from like being on a website with like a landing page revolutionizing finance for, you know, the digital age or whatever it's going to say and bringing that onto social media where people actually are and and like showing your product's use case and and bringing, you know, and finding the users that you would want to use it.

Like instead of having the slot bot replying to everything I say, what about like going and identifying people who actually want what you're selling um and then go and try and help them to arrive at that conclusion. So people like, "Oh, I've got all these tokens. Like what do I do with them?" Like, "Oh, hey, here's a bot for that, you know?" Or people asking questions on Reddit like, "Oh, here's the solution for that." Um and so I think that this needs to not just be like AI agents is the product, but that AI agent is kind of part of a deep it's like the the interface to a product. Um and I'd like to see a lot more of that.

I think I couldn't agree couldn't agree more, sure. And I guess my question is, um why do you think Eliza, um the framework itself or maybe even like the platform that you guys are building now is best suited to kind of be that home space for developers and builders to kind of, you know, answer that question like, what do people need? What can I build to help bridge that gap, to help solve something that someone needs, um compared to like maybe other platforms and other frameworks that other teams are working on. You know, you've got like I think there's these different frameworks.

There's one in Python from the Cerebro team. There's one in uh Rust from the Arc team. Like, what's um Elijah's uh special sauce per se? Um I think that I think language is actually part of it, not the whole story, but it is relevant. Like um there are more JavaScript developers building applications than anything else. We're We're currently communicating in a podcast app that's built on JavaScript. Pretty much every communication app you have from like um uh Discord to Microsoft Teams is also built in JavaScript or like some sort of native runtime with the UI and the actual interaction stuff built in JavaScript. A lot of the back end. And there are more developers who have who

are building on JavaScript and TypeScript than pretty much all the other languages combined. It's like just wildly popular. Um especially with the rise of things like React Native um and uh and other kinds of ways of like deploying these apps. Like Tauri is a good example. Um and so uh I think there's just like a large developer base, especially anyone who's like built on EVM before, um has downloaded Node.js and run Forge or Truffle or whatever these and is like familiar with the ecosystem pretty well. Um and And so I think you're just like tapping something that if you've ever made a website before, you can also make an agent. Um Python's not like a particularly hard language, but it's not super like easy to package into um different form factors, like running it

in the browser, running it embedded. And it's For a lot of people get stuck just installing Python. Like the the ecosystem is a [ __ ] mess. The package installer situation is crazy. And it could just be as simple as like people are like, I don't know, I can't get the right version to work. Um although I think Python is also like a good way to build back end stuff. But generally I I I used to build a lot in Python. I built all my agent stuff in Python. And I just came across like there are some really technical reasons why it's not great. Like it's just kind of bad at asynchronous programming. And like you have to like you just the way you manage async and it's kind of stupid.

Um it's not great at string handling cuz it uses like tabs and or spaces for for the way that you structure code. And I just found it like really annoying. And um and I like I was kind of turned on by by another friend who is doing a lot of agent development in TypeScript. And I just realized like, well, that's obviously the way. Um and so that's like part of it. Um the other the other part is that like we were just giving you an end-to-end thing. When you clone it, it just works. And I think Arc is a cool project, but there are no connectors for it. There's no social connectors. And there's other projects that are like like like ZeroBro is is cool. ZeroPi is cool cuz like it is kind of just like, you know, it's just social connectors basically.

And like loops to to reply to stuff. Um and then there's like a lot of other projects that are kind of just a lark where it's just like, cool, you got some ChatGPT talking to itself. You made like three agents chat with each other in like your command line, but does it actually connect to social media? Does it Does it connect to anywhere? And I really think um like what these frameworks are is kind of the body to the brain. Like definitely the LLMs are are a powerful brain. And what we're really building is like the body. Like the the thing that actually bridges that over to these like different clients and stuff. Um and just by providing all of that in a box, um I think we're like dramatically lowering the barrier to entry as well as like the amount of code that anyone needs to write to deploy a

product. And then they can just focus on like their product. Make a character, pull in your APIs. Um we have a like really simple abstractions for like input, output, you know, input is called like a provider. And that could be anything you want. You could get the news, you You get relevant memories, you could get like kind of your like Zach XBT or sorry, your AI XBT style APIs, um you know, of like market intelligence and stuff that's relevant to whatever someone's talking about. Go like look up the coin, pull all the data down. And then you have um another Yeah.

Hey Charles, I I uh So sorry to interrupt. It It This is This is fascinating. I I guess like, you know, I'm not a developer, right? But every time I see a a tweet about Eliza, I see like, you know, hundreds of new features being released, you know, you ask for things live. Like I think you just asked for um all the agents to have voice through 11 Labs or something recently. Like the the dev flow and the amount of features being released released is ridiculous, right? It's It's very high. But as a non-developer, I don't really understand what that means, right? Like if I have a killer model, do I just sort of bring it to Eliza and this thing is live on Twitter or Can Can you maybe bridge that gap for me? If I'm an agent builder and I'm approaching either, you know, Eliza

or a competing platform, like, you know, what do I get in in maybe non-dev speak? Yeah, I mean, um you pull the code down onto your computer, you modify your character, and you start. And you're kind of out of the box going to have a bot that can that can shill and do whatever, you know, and like talk about whatever you want. Um and that's kind of the minimum. We have a lot of plugins. So, let's say you want to add a wallet. Well, you just enable that plugin, add like a private key for any of the EVM chains, whatever chain you want, and you've got that. You can uh add API keys like like or um for Discord, or you can add your your um Twitter like you know, username, email, and that's all set up. Telegram or um some of the other sort of like places

you can connect. And that's all like kind of out of the box ready to go. You don't have to write any code, and that just works. Um which is probably why you see so many bots just like shilling and replying. Um and then and then from there, you have these abstractions where you can you know, like just just kind of a pattern that you can follow to bring stuff in or have the agent do stuff, which are called like actions, right? So, you want to uh uh I don't know, have the agent um go order you a pizza, for example. Well, you just have like a a order pizza action. But, then they're like, "Okay, well, now we actually got to get the information about the user." So, you'd probably have like a um current user information provider, and then you'd have another thing called an evaluator,

which like extracts the information that you're asking from the user. Like, "Oh, they don't have I don't know what their name yet or their address." Um and so, if somebody's like DMing you for a pizza, um you can kind of use these three different things. Like, the uh the you you'd make a provider to get the information about the user that you already have, the evaluator to extract more information, and and then the action to actually go and like order the pizza. And the agent's smart enough to be like, "Oh, okay, cool. Well, you haven't given me your address yet, so I can't call the order pizza action." Um and from those three things, those are like the three building blocks you can use to build pretty complex applications. Anything that like you see on a website that has like a form that you type stuff in, anything that's like

a basic action. Obviously, we're using this for things like the auto LP manager and stuff. And like, that's all just like, you know, it's no harder than programming any other website. It's mostly just like calling APIs and stuff that any developer is probably pretty used to. Um definitely for a non-developer, I think um you know, you're you're probably going to want to lean on like a platform that already hosts it for you, and then you'd be selecting like capabilities or or plugins that you want to add. Um and you wouldn't be going so deep into the actual code. Um but but definitely you can. I think that the the Yeah.

Sorry about that. No, no, this is this is really helpful. I guess maybe just like maybe just to ground it and and paint a picture. Like, if I'm a developer, like, how long would it take me to build that stuff myself or or piece together those pieces versus using Eliza? Like, is it a couple of days or is this a couple of months? I'm just wondering how much faster the time to market is using Eliza versus not. it really depends on what you're trying to do. Um I'd say that like, if you were really if you were just looking at the code base and you're like, okay, well, I can see the abstractions and stuff. You could probably build something that was very very specific to what you wanted to build in a in a pretty short period of time. Like I could probably have that

together in a week and feel pretty good about like an agent that did the thing that I wanted to do. But um but if you wanted to have like memories, you wanted to extract information, you know, you're going to start getting like wanting to build more of a framework that enables that. And I think that same app like I could you know, I built the pizza delivery thing and handed it to Ropy and I spent like 5 hours and he spent 2 hours on it. And you know, it's like basically a day at work. To build a freaking pizza like that definitely would have taken me a couple of weeks. Um although everything is very AI accelerated now by um how we like write code and stuff. Um but but generally the framework like provides a lot of the stuff for you. Uh I mean it's

kind of like a good example is like React. Like all website all apps are built on React. And you could totally just hack a website together. But as the thing itself gets more complex, it becomes much much much harder. So when you're doing something really simple, you can kind of just write the code like, okay, I just need one LLM, I need one blockchain, I need uh one loop. I'm going to just use Twitter. Like if I just did all that stuff, that's probably like a couple of days of work. Um but you know, we support every single model. It can run fully local. It also does like transcription. You can send audio files into Discord and it transcribes them. You can upload PDFs. You can chat with it about them. And that's all just like built in like most people aren't even using 80% of the stuff that's

there. Um so if you're looking for a purpose-built, like yeah, you could definitely just build a little chat interface. But if your goal is to um like build a full-fledged agent that can do a lot of stuff, then you definitely would probably want to use a framework that already handles a lot of that. And I and I can tell you I spent many months building this, you know, many many months. Wow. That makes sense. I mean I I'm just kind of like looking at some of the stats for the Eliza framework. And I think you guys have like over something crazy like over 48 plugins right now and there's so many, you know, get pushers to like launch even more. It kind of sounds to me sure that the if I were to summarize the advantages yeah, it's easy to spin up an agent on Eliza but also it's

main advantage is being able to build custom or purpose-built agents that are kind of unique that have this kind of extra little sauce in platforms won't be able to enable, right? And then I kind of like look at like some of the other agent platforms being launched and there advantages that they have, right? which is like easy quick design deploy no code kind of stuff coming out. So is that a fair comparison do you think? Do you think like Eliza is like kind of like best suited for like some of the custom unique built agents or what do you think?

Yeah, I mean you could take the entire arc entire rig I mean sorry you know the entire zero pi and the entire game framework and you could take all of that and put that in like code like lines of code wise is still significantly less code than Eliza is because there are so many different things in there and even just taking the plugins out just the core is like doing a lot of stuff like it's got speech text text to speech transcription PDF handling image image recognition all that stuff is just baked in and and so it's probably like pretty overkill for some people but but definitely like it enables like a very wide array of things to happen which is why people are doing so I think pulling this off and doing so many things like you know there's

there's agents that are like their entire thing is like Eliza plus well they used our pump fun action you know or like you know Eliza plus they like oh they generate images and video and stuff and and like that's all actually just built in and I'd love to see more like somebody actually posted like what happens if you just turned all the plugins on at once. Yeah, and my goal is you know, I think that eventually the agent will be able to write and do plugins from scratch cuz there'll be so much example of existing plugins that are somewhat similar. And this is all getting trained into the model. Like part of this is once you pass 100 stars and you get a certain like like certain thresholds on your code base, all of the companies like OpenAI and Claude are actually like scraping it and training

it in. And this is part of our loop is that eventually you'll just be able to write in new stuff. Hang on a second. Hang on a second. So so theoretically, if the Eliza GitHub repo goes over a certain threshold of stars or forks or get pushes, whatever you just mentioned, then all the top LLM providers kind of like all top companies start kind of scraping and using like the Eliza activity is kind of like a good Dude, that's insane. Yeah, that's part of the It becomes like an open source resource. Exactly.

Especially if it's got good documentation, that all gets scraped, too. And we put it all inside a single repo, so like the entire thing gets pulled in. And and that's like definitely a big part of what I was thinking when we're like open source this thing is that it's it's you know, if people are like, "Well, it's an LLM wrapper." It's like, "Dude, that LLM is trained on our code base." Like you do you not see where this is going? Um so Yeah. Yeah I I I also kind of like think about this like if you guys uh Let's kind of like say that Eliza becomes the richest repo. And I don't mean by wealth, but I mean by quality of enablement for any kind of agent developer.

I feel like you guys are probably best primed to attract the types of developers that aren't just crypto native, but may come from a more traditional AI and ML background, right? Is that something you're seeing or or is it like primarily all crypto crypto like tell me tell us more. No, I feel like we're really breaking through. Um so there's nothing about Eliza that is like uh a crypto project other than it has a lot of blockchain integrations, but those are all plugins. Um I think that I've noticed that the trending on GitHub thing is actually been helping us a lot to break through to the web two people.

Like people are coming over and be like, "Oh, I'm just using this agent framework for you know, stuff cuz it just seems like the thing that people use." Um and and I'm personally just very interested in uh sort of normalizing I like I just think it's ridiculous to be anti-crypto when it's very obvious to me that like 99% of agents are going to trade 99.9% of tokens uh in the future and like like that's just like crypto is the native token of agents. You can't give a you know, try giving it a PayPal account. Like I dare you. It's it's not exactly easy. Um it's a where we can just like turn on a wallet and generate a private key and you're good. Um so I definitely like my you know, unfortunately the the skill level of the vast majority of like the D gens is not

exactly super high, but there are some really really talented devs in there who are like serious builders who I think our project has really a talent like attracted. Um and and then inside of the research community like um there are a lot of people who I think like crypto, but are sort of afraid of it of like being labeled a crypto guy or whatever. There's like a lot of stigma especially like in AI. Um but like the alternative for them is to either like ask for a grant or um or raise money from venture capital. And like a lot of people just struggle with that stuff and I think once they realize like wait, you can just build things and people support it, um it really changes their opinion. Um so yeah, we've definitely been getting some traction from the the non-crypto folks, but it's but much more the people

who are like aren't actively D gen traders, but are just like, "Yeah, crypto's fine, but like this agent stuff, this is really cool." Um you know, the people aren't aren't like actively hating on it, but and are open to it. Um they just want to see it actually bring value. And I think it's totally reasonable that like a lot of people just see like shilling and hype and vaporware and are kind of turned off by that. And I think the more we show them that like this is real stuff and like um, you know, there's there's real research behind it, there's real engineering behind it, the more they like they warm up to the whole idea. So, um, definitely not as many people as I'd like, but like it's definitely Hey Shawn.

No, no, this this is a huge differentiator. I mean, one of our main reasons for us to use researchers and their ability to bring in traditional, you know, AI devs from San Francisco, the fact that you're doing this, you know, also and and at scale on GitHub is just incredible. I maybe like one question for you is maybe just going back a bit to my original question on the differentiation on the model side. Like you're now creating a community where people are creating, you know, hundreds, thousands of agents, all differentiated, all, you know, while also furthering the Eliza framework, all the plugins, etc., right? So, you're on this very cool arc and trajectory forward and upward, right? I guess my question for you is eventually you'll be pitted against OpenAI and traditional AI

labs in terms of what these agents can do and what their agents can do, right? And so, I guess my question for you is how do you view yourself competing with OpenAI's next and next model after that? Is it this swarm of agents built on Eliza working together? Is that the differentiator or is, you know, the comparison I'm offering you just, you know, not a good one and doesn't make sense? No, I think it makes a lot of sense. I'll I'll There's a few things. One, like when you launch Eliza by default, it launches a News model, um, which is a fine-tuned Llama model that like the Harmony's model, which has been trained by News Research. Um, and and I really like what they're doing.

Um, we we definitely follow them. I I talk to them a lot. So, a couple of their guys are big supporters. Um, and someone like Rob Peredo, who works at News Research and also is an agent dev in the kind of our community. Um, and they all obviously helped like launch the the god and Satan bots and and T and some of the other bots in the space. So I think that like we could say okay, yeah, we're going to do the models, but we also have partners like that who are incredibly you know, incredibly like far along with with all that stuff on their own. And and I just like I instead of competing with them, I'd rather like ally with them and kind of fill out the other side of that. That's like a really good compliment to what they're doing.

Um I I like I think that people don't I think that that's the one side. People obviously say like well do you train your own models? And it's like I think if people realized how trivial it was to train a model, it's literally one line like if I go to together.ai, I can take one command line and point it at one JSON file and I can start fine-tuning a llama model within like five minutes. It's really really really simple now. And there's like a whole lot of great tune tooling for making this a thing. People do not understand that.

That is not Yeah, it's shockingly simple. Now what Nuke does like is actually not like some crazy fine-tuning recipe. It's actually data. Like they are collecting and curating data and and that's like their secret sauce is that you can't just get the data. You it really is quite a lot of effort to to collect it, prepare it, clean it. And they're focusing on a different kind of data than like OpenAI. And this is really our market differentiator for and theirs as well. Other reason we use their models is because they don't refuse a lot of requests like the OpenAI does. And we have a word for this which is the OpenAI models are cooked.

Um that's what all the It's kind of dirty, but all the agent devs basically are like, yeah, OpenAI is cooked. Um And so like what our market differentiator is is that OpenAI's never going to let you like they're never going to make an agent framework that connects to Twitter. They're They're going to let you just like make waifus. Um they're never going to let you make your assistant super sexy. Um or like whatever it is that you want. Like they're just not edgy, they're not cool, and they have a lot of pressure on them to be like I mean look, you if you go to ChatGPT right now and you're like, "In the 2024 election, who should I voted for?" Well, now it'll probably give you some long response, but for a long time it would just say Biden. Like not because you know, because that's just how it's

trained. Like that's just the pressure that's been on it. Um and I'm not saying I'm like politically for one side or the other, but I think it's kind of stupid that like a leading model would have a political choice that's like right or left in this kind of time. Um and and I think that they're just like extremely extremely playing it safe. Um and and they're going to lose because of that. Like obviously for the average user to just get an answer to like how to make a recipe or something, like I use ChatGPT. I think it's great. But I think that there is for a lot of the products that people want to build, a lot of things that are going to make money, like they're just going to either refuse you, or they're not going to give you like exactly what you want. And so I

think that we're going to see a like now that fine-tuning is actually pretty trivial, I think it's much more about like what is the data that you're collecting, where are you collecting it from. Um and and I think like if you want to make something that the degens like, just go scrape degen data and fine-tune it. And I don't think you're going to get that from OpenAI. They've They're definitely talking about like like I don't know if you saw Sam Altman's tweets. He's like basically one of the top things that people want is like an adult mode. Not like adult like NSFW, but like like adults in the room, like stop treating me like a child who can't see this information. Um you know, and and I think that um by being centralized, there's a lot of political pressure on them, a lot of government pressure on them to be a

certain way. Um and I think the open-source movement is just free of that. I think yeah, there's a lot of fear, like, "Oh, what if we all have virtual girlfriends?" I'm like, "Just [ __ ] grow up." Like Like who That's the least of our problems right now. Um I think it's it's much better to have like high variance and a lots of different models that can kind of do what users actually want. Like just give people what they want and stop trying to control their user behavior and it's just going to win. Like you know, you're not going to see an agent from God, this No, you're I'm with you.

I totally agree with you. I mean just So maybe just a maybe to play devil's advocate or maybe just to push back cuz I'm very curious your take. Um Open AI obviously has you know, obscene amounts of funding. They're worth you know, I think 180 billion. They have a lot of talent, you know, minus a lot of founders and and key people who have left. Um and they also have the lead on reasoning and and all of these things. Um but crypto AI offers us, you know, a community overnight, incentives overnight, capital overnight and you know, limitless sort of innovation. You don't have to wait for you know, GPUs or anything else. You could just go and do what you do, right?

Mhm. Um what path do you think gets us to AGI faster? The crypto path or the open AI path? really one or the other. I think it's kind of all all of it working together. And I think this is this is like the silly thing of the rapper thing is like if if the biggest companies in the world are all doing something, then why is us competing with them going to accelerate it? I think that there's a whole other space. I think AI agents are just like fundamentally the kind of the stepchild of the AI world. Partially because they're they're not really benchmarkable as easily. It's pretty hard for like a a PhD researcher to say like, oh, this agent has scored more on the blah blah blah without like significant um like subjective kind of effort and engineering. It's much easier to like

create a new mathematical model or say like, hey, we've got this new attention mechanism. And and the researchers are really really good at math. Like you know, like on a level that's kind of hard to comprehend. Um and and and so like there the amount of code that you actually goes into like like training GPT-4. Like if you actually look at the code, it's not a lot of code. It's just like crazy math that's like conceptually very like very like requires, you know, like easily like 6 to 8 years of like seriously just studying the math and computer science and all the other papers and all this stuff.

Um and so there's kind of they're not super incentivized or they haven't been in in like academia to pursue this more like agentic side, which is much more just like basic engineering um of like you know, just just sitting there and interacting with agents and like seeing how they respond and solving the problems. It's much more creative kind of endeavor. Um and so a lot of people who have gravitated towards agents are either like publishing research that's kind of meh, like, "Oh, we made this stock trading simulation." You look and it's just like a bunch of bots with a system prompt like, "You are a stock trader." Um which I just don't think is the thing. Um or just like a lot of really schizo people, frankly, um who are like, you know, "Oh, the thing is sentient, you know?" And like, "Okay.

All right." Um W- w- w- hang on. Wait. Speaking speaking of sentience, sure. All right. This is This is probably a good a good way to pivot. Um you've mentioned many times on X that the goal is open-source AGI, right? And you've kind of like been saying that for a while now. W- W- What do you mean by that? Like like how do you see this manifesting? Is it just a bunch of agents autonomously working together and the sum of their outputs is this crazy intelligent thing or or do you like what is it? Like shill us. Shill us that vision.

Right. Right. Right. I think it's I think that centralization is part of decentralization, if that makes sense. Like all those players are throwing stuff into this larger pool. And they're really contributing the model side of this. But what they're not contributing is like all of the possible variations of what the agents actually do, like what the models actually do action-wise. Um and so there's this kind of loop that we need. We need to like have agents out there doing stuff, collecting data, ideally putting that into some sort of public repository where it can then be trained into the next models, which then enable more capability, which we then use to do more stuff, which collects more data. It needs a few things. It needs like big models which are trained on lots of GPUs right now. Hopefully that gets smaller and smaller. And if

you see there's there's like Deep Seek came out and has been scoring extremely well in benchmarks. And that was on like $6 million worth of GPUs, way less than than GPT-4. So clearly like like China not having access to all the best Nvidia stuff has like kind of been a forcing function to make them make open source models that actually require a lot less resources. And and still like outperform a lot of the stuff that requires a lot more resources. So definitely like better models, more efficient models is part of it. The other part of it is just like market intelligence. Like if if a million developers are using this and for every single product and a lot of it's open source, and then they're like competing with each other, they're kind of like optimizing the

whole thing and like all of the capability. I would say that AGI is the internet at some level. Like when the internet is just full of agents and they kind of do everything, it's not going to be one monolithic system. We could call a monolithic system AGI if we want. I guess it kind of depends on what your like what your definition is. I think when most people think of AGI what they're really thinking of is human level intelligence. Can this agent that I'm interacting with basically do everything that a human can? And I think that that there's a lot of ways that could happen. A lot of it is that the the model doesn't have to have like all this baked into it. It just knows how to like call the API where it can get it. Or it knows how it has to know how to like

operate the computer to go to the website where the information is. And so I think that if it has like general capability to use your computer in the same way you do, it has like really good memory system, It has a lot of like just functionality that you'd expect. And then eventually it's tying into like an actual robot that walks around, which is obviously coming. You're going to be like, "Okay, well, damn, that thing is like definitely AGI, like for sure." Although like I think we'll probably somebody you know in in in AI land people have kind of said like AGI is the thing that computers can't do yet, if that makes sense, that the goal posts keep on shifting every time we sort of launch the next model. And there's also this similar idea of ASI, which is

artificial superintelligence. So AGI is artificial general intelligence, ASI is artificial superintelligence. And and superintelligence would be like this idea there's this kind of crazy god-like model that can manipulate people and like take over the world and stuff. And I think that like maybe that would be possible if there were just like Microsoft building this, but if there's so many different players and they're like a lot of them are open-sourcing their models and there's so many different variations and we're taking those and fine-tuning them and doing all this stuff. It's much more like the overall swarm of the internet and agents starting to interact with each other, being able to pay each other, each having their specialty, each having like this highly variant thing. That's going to look like superintelligence to me.

That's that's like a crazy system or system of systems. And and I think that if any one of them tried to like hack all the other ones, it'd be very very difficult because none of them really have that much more capability than the other. And and while all of the companies what you what you kind of see is that we're hitting this kind of energy limit where like the models can't get that much bigger without requiring literal nuclear reactors, which is why companies like Microsoft are investing in nuclear reactors. And and so they're all just kind of incrementally improving one over the other. And you'll see something like 03 which which is definitely like basically human-level intelligent on almost everything.

That's the the new opening eye model that's coming out. And like but by the time that's out there's also like a lot of other people working on this. I mean even news has their own like reasoning engine that in you know everyone's like looking at the research and implementing this stuff and and also like something like 03 is it's going to come out and it's basically like AGI. I'd say it's pretty fair to say like when people think of like a reasoning model it's pretty freaking good. But also like OpenAI has this problem where they have like almost 200 million users and their product is they're just trying to race to the smallest thing to serve all these users because they're just getting clobbered by requests. You know their GPUs are all running hot like 24/7.

I don't know if you've used Claude recently but like they're kind of saying like due to volume we've made all of the responses concise and they're starting to sort of like do the same like being forced in the same thing. Quality degrades. It does. And so like the the research is definitely outpacing the the the actual product because it's like Sora. Like Sora was announced last year and they only recently released it and that's because they just it's not that it doesn't work it's that they don't have the GPUs to serve it at scale. And so they have like kind of an innovator's dilemma of like they can either give a few people the best thing and that's kind of what they're doing now now OpenAI just released like a $200 a month subscription. So that's kind of what they're doing. But for the most

part what they've been doing is just scaling the models down. Like can we get an 8 billion parameter model to do what we were getting 175 billion parameter models to do. And certainly like the GPT-4o mini is better than GPT-3 even though it's 1/20 the size and it costs 1/1000 this much. So I think that there's there's I don't know I think it's going to be a tendency of like evolution through like people competing with each other kind of see what we're seeing on Twitter is like everyone's trying to one-up up other on capability to sort of like one up the market cap.

The model company's being forced to make their models much more efficient. Many model companies competing with each other on like the metrics for their best in class models. And all of this together and then also like they all have a mind share problem. Like they're all trying to get developers and so they're all trying to open source and and do that which is you know Grok is is at least some versions of Grok are open source. Gemini's like some of their things are open source or at least they said they would be and and obviously meta is just like crushing it. Like meta came out of nowhere and it's just like llama is like the thing that everyone uses for all their research now. And so that that to me like that whole combination of everything is kind of where we're getting to like super intelligence like

you know and and I what I want to see is a world where like I just go on on Twitter and I'm like I need a bot that does this and it's like bam there's like three bots that do that perfectly and I can just choose one. Not this like slop I think like that's the that's the world that we're that we're working towards. I think right now we're in slop [ __ ] world to to use your word but or to use your term but we will move towards something that's a lot more seamless and slick like that. But of course you know a lot of people have to kind of like buy into that vision and and kind of like what you just laid out just now are the moving cogs that are going to get us there essentially.

I kind of like think about how crypto plays into all of this Shaw and you know you mentioned that um you were working on the Eliza framework or V1 of it before you kind of like got into the whole AI 16Z stuff and there was a token around it and it kind of got me thinking well I think one of crypto's best known use cases is it's able to aggregate attention and capital towards things right? And this can be good and and can be bad. Like mainstream media likes to portray it as like a very bad thing. Like people lose money and all that kind of stuff, but they don't highlight the good thing, which is like capital can help build new zero-to-one innovation,

right? It can help fund teams that were otherwise, you know, under-capitalized. I see you. Almost every other tweet is, "Hey, I'll give a $1,000 bounty to whoever builds this and that." And like that kind of stuff is like very, very underrated. I'm curious like, you know, what how are you thinking like the evolution of the a16z token is going to play into all of this, right? Cuz you know, you have this token and it was initially meant to solely be this kind of like governance DAO situation and, you know, perhaps be leveraged by token holders to kind of influence a automated fund bot AI, if you like, or trading fund bot AI, called Degen Spartan AI or

or PM AI, which is like a AI version of Marc Andreessen. Um but now you guys have announced this platform, right? And the suggestion there is there's going to be a baked-in tokenomics, which will drive value of like agent launchers and stuff back to the token. So it's kind of becoming this all-encompassing value capture token. How do you think about the role of like tokens in crypto playing into like all the amazing stuff that you've you've spoken about that that will hopefully become true. Oh, yeah. I mean, I mean, if you think about like intelligences, you know, I think that the market is a form of intelligence. Like it identifies opportunity, applies that capital to it.

People compete on it. And it basically optimizes down to like the best version of things. And then there's five of those and they're all competing until it, you know, ideally becomes like a fully formed kind of thing. And I really think of like a lot of, um, like a big part of the equation here is applying market intelligence and competition to things. Um, so I I think that uh, I just I think crypto is absolutely part of of it because it creates this like like it's two things. One, it crowd funds projects that um are driven by what people actually want and not by what like some VCs in San Francisco want or think is valuable. Although they have a lot of insight and good opinions, a lot of times they also just drive stuff into

like we'll invest in founders. It's like, do you mean the founders who live down the street from you or do you mean founders? Um and and I think that it's like it's very much like a Silicon Valley bubble which sort of like creates some value but also like prevents a lot of the more decentralized just allocation of capital to the right people. Um and and and there's sort of like a um like the the hype is sort of like what do people actually want? Like what emotionally gets them? And if you could deliver that to them, a lot of times they're going to be extremely psyched.

And the main problem with crypto is that people have hit that emotional tone but then haven't delivered, right? And if they did deliver, we would most of the things that people are throwing money at, I think people will be super excited by. Like if there is actually a bot that gives you like perfect market insights, like yeah, that's that's worth a lot of money that's worth infinite money really. Um and um it's just that like sometimes it falls short. Um but I do think that as more serious people, especially more of this like uh sort of auditable stuff. Like open source is auditable. Anyone who actually knows how to code can go and like read the stuff and be like, okay, that's legit. Um and and when you can identify like the legit opportunities and those opportunities speak to what you really

want, um then then like it just gives you a way to directly get exposure and to invest in those things. I think one of the frustrating things about the world as it is is that like you just can't invest in Open AI. You can't invest in most of these companies unless they go public, but by then it's like, you know, yeah, you're going to see what like a 10% return over 6 months or something, you know, like okay, cool, their market's going up, but it's not going to be like like, yeah, I got in early on this thing and you know, now I have generational wealth because I funded the future, which I think is like the ideal dream for a lot of people um in crypto. So, I I definitely think if we could kind of like figure out better ways to prevent scammers and

prevent um like stuff, you know, and and this is where I think open source and having open development, people like streaming and being open about what they're building, um then then it kind of enables like a lot more efficient market like capital allocation than I think we're seeing. Um and I think that's just going to massively accelerate all of this stuff. So, um and also like it clearly the agents are going to trade tokens with each other and we're going to tokenize everything. We can tokenize trust, we can tokenize merit, we can tokenize money, obviously, um and and like yeah, I think that's like it just seems like like capital yeah, sorry.

Hey Shah. This is a No, no, I'm totally with you. No, don't No, this is amazing. I mean, I'm with you, right? It's impossible to invest open AI. I've even tried to do the SPVs and and man is it hard. I haven't been successful there yet. But I guess my my more direct question for you is like the crypto crew saw the Cosmos ecosystem play out, right? Like awesome code, like very good community, but people forked it and they built ThorChain, right? I think DYDX and like many other killer projects with with very limited or if not, you know, mostly zero value flow at all back to, you know, Adam and things like that, right?

Um so, you guys launched your token plans and and you're planning to launch plaid, you're you're planning launch fees and LP pool pairings, you're you're planning fee sharing from the launchpad and buybacks and burns and and there's a lot of ideas sort of floating around. Um so, I guess my question for you is like thinking about the velocity of those changes, do you think that AI 16Z can implement these value cap mechanisms before say an agent, you know, steals the show and and then is off the platform or another platform takes the cake. Like, I'm just curious how you're viewing uh the speed of things there and the big problem with uh open source blockchains is that there's a lot of incentive to fork because there's a

direct monetary incentive when you own the network token. And I think if we launch an L1, then everyone's going to fork that L1 or they're going to feel like they can't really work with us because we're an L1 and they have an L1. Um and I think if you look at like how much tribalism there is in uh in crypto, and it really is driven by the fact that like it's kind of either or um as opposed to being like both and. And um I think that like I I'm my my realistic take on our tokenomics is that we're going to have to continue to evolve it and find new ways to make money. Um I think I we have a lot of ideas there. I think the the goal of the launchpad is not to be the tokenomics, but to be kind of the V1, which is to say like we have

a lot of attention on us, a lot of partners already want to launch on our platform, and they just want like a hosted way to uh host their agent. Um they just want to do like the pump fun agent thing and then go and actually add the capabilities. And we can go and give them like all the plugins, all the capability of our ecosystem, and they can just like, you know, use that. But my guess, um and look, we're going to open source this. I want to open source the actual launcher. Um and and then everyone's going to copy the launcher. And then any company whose entire tokenomics, hint hint, is just a launcher, is really going to have to rethink what they're doing and like what the bigger picture is cuz probably just putting up character slot bots is like not going to last much longer even if

they have like a a kind of burn and buyback sort of strategy or whatever it is. Um so, I think like in the long run, what we're really looking to do is actually invest in technology that grows the pie. And I think that we have some ideas for how to do that. But the near term is that we do need to like just launch what people want. Like a lot of people want a launchpad, let's give it to them. I think in 3 months, it's going to be like, "Okay, cool. Launchpad's out. Everyone's got a launchpad. Whatever." Um not that hard to build. Uh there's so many tokens that like most of these agents are probably going to go to zero. Um but some of them are not. Some of them are going to be clearly valuable as the teams actually embrace it and and

like build real stuff on it. And then like what's next? Um I think what's next is like um not just launching uh agents, but actually investing in projects that make obvious value. And we're already making investments. We're already doing acquisitions. Um and those things also have like their own tokenomics where like we, you know, get like earn revenue, take that revenue, and buy back our token, and use it for more investments. Um and uh and so I think there's like a lot of ways to create buy pressure other than just like earning a network fee or making people buy your token to burn it for another token or you know, whatever the or or parrot or whatever. I think those are like really good and simple and like easy to implement things that we're definitely

going to do. But I also want to like push us through that meta onto something bigger. I really I really think that like um what I want to see is like legit the A16Z plan of like, "Let's actually go grow the pie. Let's find new technology that's valuable. Let's pre-validate it with attention." Um and so we're building kind of like a a production studio where we can have people pitch their projects to the DAO and to the the characters. Um and then the stuff that's actually hot and that's getting hype, we can then go and make investments into.

And I think that um so I I like our tokenomics plan as it is, and I think that's going to last us for, I don't know, 6 months. But I'm also like also looking at the next tokenomics plan, like the next version of this thing. Um which is like, yeah. Stop. This is a This is an incredible answer. I mean, the the idea that you don't want to be an L2 because other L1s and L2s won't work with you is sort of extremely contrary to others' uh point of views. Like, look how many L2s we have but it just makes so much sense right for agents to work cross chain and and to work with you.

Maybe a a further question for you though if you do now the token econ and there is value current AI 16Z the token and you know not financial advice here but if it works out there is like a pretty synergistic flywheel here where you have more money as a project to develop the platform but you also have agents that are furthering the open source framework and building it out sort of on your behalf in a in a non-directed way so you have this direct you know growth of the eco then you have this indirect growth that sort of feels pretty crazy to me like do you do you think about that growth like that cuz that's that's not really happening now right?

I think about it a lot. I mean so there's this idea okay so so in in AI there's this idea of foom and the idea of foom is that at some point the agents are actually able to write their own code and then they just start writing it faster than any human can and then they're just looking for like every single possible use case they can write code for and then submitting it as a pull request and then other agents are reviewing it testing it and like that could happen in a couple of years like that could happen in under two years and so there's definitely like a longer play here if we can just get far enough like it'll hit kind of an exit velocity where like the thing just goes way faster way like in exponentially faster and then we're kind of in like okay this

is legit AGI like this thing is building itself you know so I think that like whatever we can do to accelerate toward that reality which seems really inevitable to me I mean it's already like there there's like projects like Soul Engine reality spiral which are submitting their PR's to the GitHub repo they've already been agents submitting pull requests and so yeah it's like can we if we can accrue value to the token and we can invest in our ecosystem to grow those kinds of things which then grow the ecosystem which then get more value to the token, which then grow the ecosystem. uh Um, eventually this is going to have extra velocity and we can all just like put our hands up and let the thing run.

Um, I I do think, um, you know, there's a lot of practical stuff we have to do to get there. I do think it's really important to just have like very basic ground game fundamentals of like making sure that our token is accruing value in the meantime through things that we would expect and things that people want. Um, and like the launchpad is really just listening to what people want, what people are already building and like giving it to them. Um, but then in the future like you know, you'll be able to ask the agent like, "Actually I want you to build this thing." And then, you know, 10 agents will try and one of them will succeed and then we'll all vote and put that in and, you know, I I don't know. This could get like really crazy really fast.

You know, and that's the that's what we're trying to accelerate toward. Yeah, that's the fun part. I guess maybe switching gears a little bit from token econ, I want to talk a bit about Solana versus Base for agents because, you know, you're I think primarily focus on Solana and and your competitors or, you know, competition players are are not and focused on Base and other platforms. I guess my issue with Base is just the centralization of it and linking it to a company with shareholders who have direct opposite incentives on centralization and value flows and things like that. Um, it might be a loaded question where I'm sort of on your side here, but what is your take on what Chain AG I will eventually live on?

say that from a user perspective, we'll know that blockchain has normalized when people don't even know what chain their tokens are on. Um, obviously like the EVM and SVM models are quite different. Programming on them is quite different, like the capabilities are different, but from a user perspective they're basically the same. You know, you swap tokens, you open your wallet, you see you have money or you don't. Um, and I don't think it like I think that what I'd like to see are agents kind of like glossing over this and just sort of, um, like definitely we're going to see a lot of tokens bridge between the two. I think um we would actually already have tokens bridged to base, but because we're an SQL 2022 token with the with the mintability, it's been kind of

challenging. Um I think we're getting over that hump. Um I actually I also like the base team. I think that um you know, they they've been pretty good to us and pretty supportive. Um so I don't I don't want to say like you know, I I don't think I personally have a pick. We really went with uh Solana because that's where the users are. And I think that as a product person, we should just like put our ideals down and look for like what are people using and just give them what they want. Um and give it to them where they want. I think that uh you know, our agents like you can launch a base agent today. There's it totally works. There's no limitation. You can launch like a StarkNet agent today or whatever you whatever your thing is that you like. Um I I do think that the these

these ecosystems have really been split um by like the price of their respective token or if they have a token or not or like who the people who are already building on them are and like what the infrastructure is. A lot of why we chose Solana is just that like Dialect Fund was on Solana and the users were on Solana. Um but um but I I'm pretty I'm pretty agnostic. And I think that like the best play here is to sort of cover all the bases, just watch what people are actually using and then give it to them there. Um you know, so I I don't know. That's that's my contrarian take. Yeah. That is a No, it's a very fair answer. I mean, I have a ton of respect for the base team and and the projects building there, too.

I I definitely want the enemy or the the competitor to be Web 2, right? Not not our our Web 3 uh cohorts. Um but Shawn, maybe uh just one more I have to ask you. Um I think we all agree that these slop influencer agents are going to zero. So there's sort of this time period where they don't do well. Um and then we have performant agents doing real utility type actions and and performant things, but there may be a gap between those two. Do you view that as a soft landing? Like those go to zero and and we get the agents in a natural transition, or do you view like a pretty big time gap until we get to these really performing agents, you know, well

after people get burnt out of these influencer ones? I've seen so many things that are just like on the verge of launching that I think that we're going to see these the one meta just eat the other. I think they'll actually have this meta continuing until the next meta like just clearly blows that one away, which is going to be agents that are doing crazy stuff. For like like legit crazy stuff. Like, oh, what feature should I build? Oh, how about this? Okay, cool, I built it. Can you go test it for me? Like, you know, I don't know. What whatever or like, hey, I've got this really cool product. Or hey, these people are actually coming and making money with this. Like if people can make money off an agent, that agent's probably going to go real well, you know, from that point on. I think

that there's there's some kind of interesting like part of this is like I don't think that the slot bots necessarily go away, but what's happening with X is that they've kind of realized that like they can't really arms race away the bots. Like they can't really even tell what is a bot without human reviewing every single bot, which again is basically impossible since they are like Turing passing agents and I I can't even necessarily tell who is human or not anymore. And so, their solution has been to just make the algorithm more punishing of people who are annoying. And so, if you get blocked now on X or you get muted, that like a lot of people block you, that actually is like the biggest signal to sort of like shadow ban you. And and

the agents just won't get a lot of traction if that's the case. And I think that what we're seeing like I I just block the bots that reply if they don't have any business replying. I think if they don't if they're not summoned and they're not offering me anything, like I don't I don't really think we want that on the platform. And I think that's why we want this social agent thing to happen is because it kind of creates a self-aligning force where like we're all like we can all reject this and we can just like have a conversation about it online and then go block the bots we don't like. And then those developers are like, "Shit, [ __ ] Okay, okay, okay. I got to I got to do something. I got to up my game, you know?"

Meanwhile, all these other agents are coming out. They're doing impressive stuff. And I also think we just haven't seen the meta of like DeFi really hit agents quite yet cuz the teams are still building that stuff. But I think in the next month you'll see a lot of things. I also think we haven't seen like agents that go and find users for their product. It's a lot of just like shilling under KOLs and stuff. But imagine that, you know, you have any need and that agent's like, "Oh, cool. I've actually got the solution for that." You're not going to block that agent. You're going to be like really thankful that that agent was like basically the new Google or you know that or that social media's new Google for the product you actually wanted.

And and I think that that this stuff is like not hard. You could definitely ship something like that in a week. I think people be just need the incentive to do so. And I think that they basically like I have a friend who gave me this great analogy where he's like, "Currently we're in this space of like dogs playing poker." Like if you walk into a room you saw four dogs playing poker you'd be like, "Holy [ __ ] there's a bunch of dogs playing poker. That's incredible. Like this is the craziest thing I've ever seen." And then a couple weeks go by. Yeah, a couple weeks go by and you're like, "But are they good at playing poker? Like are they actually making money or they just like holding these cards?" You know, like what are they actually doing? And like once the novelty wears off, I think that now it

becomes like, "Okay, like who's got the best dog? Actually, you know, who's got the best poker playing dog rather than whatever." And and so I think yeah, I think that we're going to get past this like definitely influencer agents are here to stay. I think what we're going to see is actually a lot of them like in web two. Like you can imagine like McDonald's actually making like a Grimace agent and people are interacting with it. You can imagine a lot of influencers who are like posting content and then getting their DMs are smashed like building a reply bot that's actually like creating these parasocial relationships with all of their fans.

So the Mhm. Yeah, the the crazy thing too that I'm struggling with with all the agents is it's very hard to get a visualization on the the architecture, right? Like you know, disclosure I'm explore the token. I really like the founder and what they're doing in the evolution side, but when you visit that website, you can sort of view the TEEs the agents are built in the attestation so you know that it's not a robot. Again, I'm not a developer, but it it looks good to me. Um but it's it's hard to see like is there a verified inference? Is an agent on a TEE? Like what model is the agent built on? Like where is it hosted? Like all the pieces of all these agents it's very very hard to find. I'd love a site where I could just view all of these

things, right? But it's just I don't know. It's a lot of just trusting the devs at this point, right? Maybe someone will hear this and then they'll build that because I agree and I think there's probably like a lot of opportunity there. You know, TEE has been around for a long time um and I've talked to a lot of these devs and before the agent thing like they were it was they were just like a really esoteric obscure thing, right? Um and once agents came around people started saying, "Okay, but is it is can't you can't you just rug it? Can't you just like yeah, okay, it's an autonomous agent like what's stopping you from just taking the private key and stealing all the money?"

And that's when people really started to hype on TEEs. I think that you know, follow who did the the spore thing like like there I think this is they're having a very good year um because it's like created an obvious desire for this like very abstract concept of remote attestation and these like you know, these things to prove that they're they're verifiable. And I also think the same thing is happening with like ZKML and and this other stuff where it's just like, "Yeah, why do I need that?" Until at the end you know, and once you see like the product implementation uh the reason you really needed is cuz like how do I know how to trust that all of this isn't just a lark? Um and so I think we're going to see a lot of that as a reaction to like the the

uncertainty that you're feeling cuz the uncertainty you're feeling is like a great product opportunity. Um and I do think that somebody who built like a really great list that was able to like give verifications for these things. Um you know, if you go to Deck Screener and it's like got like the the trust score of you know, from from Go Plus and from these other companies. It's got the like, you know, dev sold or didn't sell or you know, little check marks and things. Um you could see you could imagine the same thing for agents. Um and I also think that like there's going to be a lot of incentive there to open source. Because if the code is relatively cheap and the main problem is trust, then like why not just open source it and let everyone look at it and see. And you know, that there'll be

like kind of a Zach XBT of developers, I think, who's like, you know, saying no, that's garbage or yeah, that's is legit. Um there could be a whole new class of influencer programmers who are like doing that. And I also think there could be products that are that are normalizing that. I think in in 5 years you'll you'll totally have like some little like, you know, you look up any agent and you you'll you'll just like be a website that tells you that. If not this year, someone should start building that for sure. I love it. Shaw, well, well, as a disclosure, I'm I'm obviously long AI 16Z, which you know, I feel good about. Um but definitely want to make the disclosure for people are aware and and I'm transparent. Um but Shaw, it's been an absolute pleasure hosting you, man.

Just watching the growth and everything you've built and the community built is just it's really awesome to see and I'm glad we were able to to host you. And Ujas, thank you so much for for co-hosting. for having me, Shaw. This was awesome. Yeah thanks guys. Lot of fun. I'm a little I'm a little uh Enjoy Soul Show. Definitely uh catch up soon. Take care.