The coming Agent to Agent Economy | Shikhai Wei | Carra Wu | Eskender Abebe | Jeff Wilser |

Transcript

[Music] okay my friends uh if you left a teammate at home or a developer back home um everything's streaming on YouTube and twitch so you don't miss a minute so don't forget to share the info um all right who's ready for the coming agent to agent economy uh please join me I hear some claps yep me too sweet um no longer tools agents are now um negotiating and collaborating and building this really interesting new economy and Marketplace so um please join me in welcoming Jeff wilser who's host of The People's AI podcast

thanks thank you so much great to be here hello I am fired up for this topic this is arguably the hottest Topic in the space right now so I have three very folks very smart folks with me to discuss this please welcome CaRu from as z a skender from Eliza labs and shakai from Asian starter so there is so much to tackle here this space is wild and growing seemingly by the day or the hour to make some sense of it and bring some struct structure I'm kind of envisioning a thre part or even three act play here act one being today the present where are you

with AI agents act two the future where is it heading and act three how do we get there what are the challenges how do we overcome them so let me start a basic question um what is the landscape the state of play right now with AI agents and more specifically decentralized crypto AI agents their space feels wild there's like a there's hundreds and hundreds of projects Cara how how would you uh how do you actually first before you men a bit about yourself and where you're you know what you're focus on then how do you think of the landscape right now of like what's working and what's not working with AI agents yeah um I can quickly introduce myself I'm Cara I'm one of the partners on the crypto investment team at a16z I

spend most of my time looking at the application layer of crypto so a couple of years ago that was games my background is in games um these days I spend a lot of time obviously with AI which is why we're here today um all the way to story which has a big presence uh today's eth Denver um where are we in terms of product Market fit for AI agents we kind of talked a little bit about this on the phone earlier I think that there are two places where agents are really working one is like agents as entertainment which I think people like to Discount I think of this category of Agents as very similar to streamers right there was a period of time when people were like streamers aren't a real profession like a being a twitch

streamer is in a career right like that's not an industry today it's abundantly clear that that's true and I think that agents specifically agents that have all sorts of capabilities for rendering their own images or streaming and interacting with their fans like that's a real industry right in the same way that any other influencer entertainment like you know game um platform is is like a is a real business um and so I think that's one area meshed with like the AI agent tokens and sort of mcoin trading as a a product that has product Market fit I think that's one area where agents are working and I think the other place where agents are working is obviously coding agents um and as most people here in the room are

familiar with like coding agents have been at least you know in in its early days like primarily centralized but there are efforts now both within web 2 and web 3 that are trying to make these you know more open source like open router is trying to make it possible for you to not be locked into your subscription right for you to basically Multiplex your subscription out to a bunch of different apis all the way to Skyfire which allows you to make agents that are not locked into like any apis right and so you can do usage based pricing for any API like bring your own API to this agent and instead of having to pay $25 per subscription per API access you can you can just pay per token like that's a very powerful thing um but the the short of it is that I

think that like crypto plus AI agents is a very nent field and there's a lot of infrastructure left to get to the point that I think a lot of people Envision which is like you have these Sovereign agents that are like your personal assistant doing things for you interacting with wallets you know interacting with other people on your behalf going on dates for you on other people's behalf right like I think there's a lot of um that's the most important use case clearly how do we get our agents dating each other let's let's build that I've talked about this for a long time as like have an AI agent trained on your personal data set go interact with another agent train on somebody else's personal data set and give me a confidence interval for How likely that data is to go well done done and done all right Shak jump in here

first give us a quick uh background on what you're focused on and how would you describe what's working and what's not in the current agent to agent economy for sure hi everyone Shai here co-contributor at well we just rebranded to virtuals Ventures uh and I'm also co-founder at lash Ventures so been investing in the space but now of course like all in on the the agentic space uh and introducing a little bit about virtuals as well in case uh people aren't aware uh is a platform to tokenize uh agents and building the uh agent Society of Agents now where we are today uh if I can relate to it from a different angle from a market perspective uh I think we've just had uh the first wave of wildfire that spread

through right and and that's purely from novelty and people like oh wow like agents can do this like AI can do that uh and we just as long as it's like done by an agent or like some level of autonomy people are like oh wow that's that's crazy and we're paying attention to that but I also think that we are that first wave is is kind of like fading now right it's starting to be like okay okay know it's kind of cool but it's still not as good and now we're starting to see okay now we need the real Embers like the the real uh utility or like usefulness to come out um and I I think that H that also applies to what you just talked about entertainment agents uh right initially people are like oh my God there's all these like personalities and like mus musicians art

that's been created uh and that novelty effect drove a lot of like speculation and got a little bit ahead of ourselves and now it's like we even get a little bit annoyed right we get these replies um and we we saw this with the flagship agent on virtuals as well with Luna but initially people were like oh my God like she has her own brain she's like creating campaigns um and now we're still doing a lot of things like Luna's doing a tarot card reading at the story protocol Booth that's super cool um but I find that people are starting to be a bit more demanding of of what these agents can do I would say the other thing that is working better right now in the same way that people are more demanding of influencers now right like if you go back and watch those old

school YouTube videos you're like this was funny but in the 2010s we were like this is the funniest thing I've ever seen yeah exactly I think that's the thing about entertainment it's a bit of like fashion right like the moment The Meta is kind of mature you need the next wave of thing coming so that that's being worked on um I would say the thing that's still retaining some edge here is one example is AI xpt I think people are still following it and they're starting to see the cracks in its armor they're like oh you're not always correct like sometimes you get the numbers wrong and you change your mind you're bullish about everything uh but it is still a good uh agent to surface timely insights and Alpha from sources that you might not be following uh on crypto Twitter so and then we starting to see a lot of

Builders trying to integrate defi integrate social and well hopefully we'll see more of those come on uh we seeing a lot of people trying to do agent to agent communication and with that I'll hand it over yeah scander give a sense of who you are what your focus on and what's working and what's not today yeah hey I'm es scander I'm the chief product officer at Eliza Labs uh at Eliza Labs we build Eliza OS this is a multi-agent framework that lets anybody anywhere create manage and deploy their own autonomous agents and then the question was like state of play what did you guys used to be called we used to be called AI 16z thank you and do we have any lawy in the in the crow okay and this emerged out of mimetic love for the real a6z where we really

wanted to like emulate kind of like the support that they provide their ecosystem to the support we could provide to like our developer ecosystem uh and then when it uh took off and really agents captured kind of like the imagination of everybody and then our IP was getting like a lot of notice we realized we had to transition and like frame ourselves closer to the actual Eliza OS framework um yeah and then so for the state of play like with agents right now I think when social agents broke out onto the scene it kind of captured like people's imaginations because they were used to like you know interacting with Claude or chat gbt and like having kind of like a very sterile experience with an llm and then like seeing these things swearing or having like consistent personality leveraging

tools and then using those as part of their personality on social media kind of like opened up people's imaginations to what they could do and then uh now it's gotten to the point where okay all the lwh fruit of like you could deploy a social agent quite quickly the novelty of this is kind of worn off and now like the harder engineering part of like how do I get the actual like memory uh to function in a way that I like how do I give this agent tools so that I can actually perform some of these tasks on my behalf with a high enough confidence interval that I'm actually going to entrust it with like a lot of money uh are starting to emerge so we kind of captured people's attention early and then the industry kind of got ahead of itself but now I think people have like

been captured and are very interested in agents and are now starting to really tackle the hard engineering problems of making agents super performance so I know that predicting two weeks from now or 10 days from now is a Fool's errand in the space that said backstage Cara mentioned that you know when you guys make investments at a16z you're looking at like 10 plus 10 12 years right so I'm wondering let's go wild what does success look like in 10 12 years sh let's start with you what would you say say what what what what do you think as a world that we' we've freaking done it we cracked the code 10 years from now agents to agent economy is booming how does that look like from the average person what's happening all right if I can try to paint the timeline of how we

go from today to the full kind of agent to agent sci-fi World um I think today we're interacting with know dozens of apps you know you open your phone you have to find an app you have to navigate the UI or you're going to different websites you're clicking buttons you're trying to swap you're trying to like use dii you're trying to use your banking app and so on I think first a lot of these will start to become more human uh level or like be human language or more intuitive just from gestures and so on uh with a voice like scarlo Johansson maybe yeah and and those will be agents right we starting to see this already you prefer to you know do Vibe coding and like hey I I can just speak to it right and learn new skills so the the barrier to access these apps and learn

new skills going to go much lower uh and this going to happen through agents now that's still human to agent interaction right um but as you interact with more and more of these agents and they become specialized in their function you start to feel like oh no now I'm interacting with too many agents and now you want to interact with fewer agents I want I want to have okay I just want one agent for like all my work stuff I want all One agent for all my um social and entertainment life I want one agent for my Personal Fitness and Nutrition and like my therapist kind of thing where I give it like very private kind of data or like uh feedback interactions and and maybe in the end we have like just that one Jarvis kind of agent that really is The Mastermind and then talks to all the

other agents and that's when we get a true agentic economy and agentic Society where like hey they actually need to interact with each other um and alongside this I would say that parallel to this we will have start to spawn autonomous communities and ecosystems as well because some of it here here this this whole area I'm talking about is like to serve humans but humans Also Serve larger purposes right we serve for example certain causes that we really believe in it could be an environmental cause a social cause a political cause and it doesn't have to be human serving those causes there could be agents that you give them these objectives and they are then autonomous Champions who can then recruit other agents other humans

to achieve these objectives and you can think of these as like zero day company uh zero person companies zero person political campaigns zero person social campaigns zero person trolls sounds great yeah exactly you can imagine a whole Army of trolls that like the whole uh well but then it depends on how much resources then that other agents and other humans decide to dedicate how much attention how much revenue then flows to these then it will that that will be the Sci-Fi future where I don't know humans will we we might have fewer humans than than agents running around in the next 10 to 12 Years love it slte it SL love it slte it scander how about you what do you see is a 10 years out yeah I think my 10e kind of like vision is like Eliza

building herself I think this is like a very good construct for all the things that are Downstream like coding agents need to have a certain level of autonomy and performance to be able to do this a lot of hard engineering problems between today and Eliza being able to build herself uh Cara from the Investor's lens how do you see this I want to talk really quickly about Eliza replicating herself and then I will answer your question have you guys heard of this concept of a vanyan probe no okay so the idea of a noyman probe is like a robot that you send out to earth like a probe that you would send out to space send a message to the other side of the Galaxy right well it's like a robot you send out to space that can then go replicate itself so it has the

capabilities of like searching for the minerals that it needs the rare earth minerals in order to create chip in order to be persistent and that's to me the first thing I think when when I think of Eliza replicating herself is like is the venman probe um from the Investor's perspective um well so I I want to talk maybe very specifically about the hard challenges that you brought up as scander which are like the the universe that we have made on the Internet is built around humans it's built in the shape of humans and us trying to retrofit AI agents who have computer use maybe right who are able to like log onto the internet and whatever interact um still needs a bit of modification in order for that to happen

so like a good example is robots will need or not robots I guess agents I'm like stuck in proband now um in 10 years they'll be here too I'm confident the robots will be joining us e Denver 2035 well they're going to be way more of them than there are us we won't be here our robot agents will be here but yes um but agents are going to need their own like decentralized identity they're going to need their own citizenship their own bank accounts right in order to be creators and get um payouts from from X they're going to need to have their own decentralized identity on some level even sign captchas right to sign a capture there's going to have to be some replacement for the capture we're going to want to be able to differentiate between low-end Bots and high-end Bots

right an agent that represents a human being that has a bank account that can post a bond right like if you're an e-commerce site and you want agents to be able to make purchases on behalf of other human beings you're going to want to validate that that agent has its own bank account and so instead of a capture maybe instead you have a oh post a bond and then we'll post a challenge to it right like so instead of a caption maybe there's like an economic process that occurs to differentiate between high and low-end agents um and so all of that I think requires some notion of or a gradient of personhood right so the like highest end it's worldcoin I've scanned your retinas and at the lowest end it's like the 100 millionth you know self-replicating bot and somewhere in

the middle is an agent that like represents Cara woo that is like dating on her behalf and has like access to one of her bank accounts like right there there's a spectrum of humanness totally and that's a great segue to what it will take to get there and I think you brought something very important which is the the authenticity how do we know what's the what's the kind of cryptographic key which brings us to crypto and I'm sure many folks in this room and watching this panel on live stream are actively working on Solutions on building infrastructure for the rails to make is happen so I guess I'm wondering both at a high level but also a little bit of like a technical level what are are the key unlocks that have to happen to make this a reality there's and there's so many different projects

doing so many different things what is it is the main obstacle here they have to kind of like figure out a way to work together have the Lego pieces snaap together or is is is it Tech that has to be built or is a tech kind of already mostly built but we're not sure how it all locks in place or is it more about regulatory issues and cultural communication and people are just level what do you see as kind of the main the main uh obstacle to get there uh go for it sir yeah I think the like biggest blocker like abstracting technology because of course you can always create better Tech is like people being able to like put trust into the behavior behind agents and I think like as long as like there's a gap there

because like agents aren't deterministic right like they're not going to be great at doing like uh some sort of financial flow where it's like you cannot make an error like they can hallucinate they have all these issues and so so that kind of reduces our ability to have confidence in them and then develop trust so as we're able to like improve our confidence levels and their performance and then build trust over time that's when I think you'll actually start to see like the maturation of Agents uh like in economies Shak what do you see is kind of the the the most the toughest nut to crack to make this a reality if I can focus on right now apps apps apps use cases go for it I know there do we need a killer app do we need

like a aside from dating obviously aside from that what is the do we need like a something that even the normies will we'll will'll latch on to absolutely absolutely I think it's a mindset shift I think in crypto we we have a certain tendency to love the the infra and it's absolutely essential and we've been doing that for so long and we still need a lot of it right we need the we need the guard rails we need a verification of like both humans and Ai and agents but right now we need those use cases we really need them uh and not just Within AG but all of crypto and I think the successful the best ones are are going to resonate first with with the crypto crowd and then also break into the mainstream so that that's my biggest push even if you're building in right

now hey show how the use case can play out right Tinker it with with yourself like go talk to your users uh unfortunately we we're still pretty isolated and I think the P pendulum has not swung enough the other way yet car you started us off down this path of of what has to happen in the future anything else you see as kind of the most the the toughest barriers to overcome um I'll I'll describe uh a a challenge that exists in crypto that I think uh like we need to work on that maybe doesn't exist in web 2 that makes the adoption of like specifically crypto agents a little bit more challenging than the adoption of agents in web 2 and um the thing I'm describing is like in

web 2 we have a beautiful system of like of of services architecture we've got all these apis that you can sort of seamlessly call and um the integration lift is like a post request like the integration lift is beautiful in web 2 in web 3 if you want to integrate a new service it's like a whole like smart contract integration process right it's bespoke it's challenging uh you can accomplish multiple hops through different smart contracts grandma has no chance basically yeah exactly you can't accomplish like multihop transaction um in one transaction you can't guarantee fault tolerance across you know say you're going cross chain or you're you know you're moving across lots of different smart contracts right

like that's different from web to and so um I think like we need to get to a point in in web 3 where the architecture is as beautiful and as simple as Services architecture in web 2 and there are some interesting companies that are working on new like composable run times like holiday um that will help enable this this but it's still early days like we are not where the API economy is smart contracts should be like apis that's like the dream right is permissionless censorship resistant apis that was what like smart contracts were is like somebody deployed a service Into Thin Air and you can like anyone can access it that's fantastic that's like what the original vision of like the you know the API that you can't deplatform

was but we're not there yet because it's much higher lift to integrate a smart contract quick I want to as we're wrapping up each of you like a 10c prediction of any time frame you like for the a agent economy a skender quick prediction um h a prediction for the agent to agent economy yeah I think the the edge for web 3 agent agent economy is like the use of cryptography and smart contracts and I think that actually moves slower than the web 2 kind of like Asian economy which is like API power thus they can move faster so I think like web 3 agent agent economy will be much smaller to start and then as those efficiencies and those architecture changes come in it'll actually start to slowly gobble up web 2

because we can offer higher levels of trust all right uh 10 seconds shy within 10 years all of us will be human agent hybrids we're going to have digital twins we're going to have human agent Cara uh agent watch dogs like government watch dogs like doji is a good example of a watchdog that is like you know constantly monitoring like the.gov web page pages to see all the things that doge is doing I think we're going to have like agent Watchdogs amazing we're out of time that was so much fun thanks a lot guys enjoyed it appreciate it thank you [Applause]