Official ETH Denver AI Summit: Agent Coordination - swarms, ecosystems and market places

[Music] great I think we'll get started with the panel um I think the easiest way to get started is just like like each panelist can give a quick introduction of themselves and also say one use case of AI and crypto which they particularly excited about can you say that one more time uh introduce yourself and one use case for AI and crypto which your particularly excited about should I start hey what's up everybody I'm Mike Anderson from think agents um we are building the erc20 of agents and what we mean by that is how do you standardize an agent so that it has an addressable name the wallet and a way to be able to network and communicate with each other uh we started the independent AI Institute about a year ago we've got 50 projects that are part of it uh some names like like render ather Morpheus like a bunch of the best players um have kind of come together to say how do we standard demand um we work very closely with futuree to get everything interoperable between all the metaverses uh and in game and so we're building that standard and would love to work with everybody awesome and then the question was uh what excit you about and I'm I'm excited about how um how crypto tokenizing these agents actually allows these agents to go into any environment awesome uh my name is James Young uh I lead a project called Collin but also working on a new project kind of a side Consortium project called mother AI uh it's a system so collab land we do token gating uh and with mother it's a it's a multi-agent coordination ecosystem and what I'm most excited about is being able to just lower the barrier for retail for people to come in because they don't know how to bridge Swap and I think AI agents because in the collab land kind of perspective we do token gating in Discord and Telegram and it's a great interface for groups to chat but also it's also a great interface for agents so we see this kind of future where you can be in telegram Discord and be able to do things without having the all of the like knowledge that you need to swap bridge and all of that hey I'm David um founding member of all us um we have I think two uh offerings one for consumers uh it's an app called pear where you can actually own your agent and that also underpins our sort of General belief we should be able to own our agents the same way we own our keys um the other uh thing we just announced today um is we have within Olas already 2.4 million transactions between agents um and we kind of uh generalize this now into a Marketplace for agents um which yeah we we um see growing a lot because agent to agent transaction is a thing um so when we come to use cases um yeah for us it always starts with uh owning your agent and I think agents in general are a great way to abstract complexity and obviously in crypto we have a ton of it so that's a great starting point but I also like these um very open-ended use cases like social agents where they can evolve their personas and actually what they do and maybe create a person it of their own uh I'm Jay uh ztz I'm at near AI near AI is uh a whole bunch of things uh our goal at the moment is to make uh make it as easy as possible for developers to build agents that have a whole bunches of resources that they can use like yesterday I think we launched a cloud operator that you could just use with one line of code in your agent uh we give away free AI inference for everyone and we will do that forever uh so if you're building an agent and want to host it somewhere host it on us or host it on anyone else here we love everyone here uh but um uh what I am really excited about as a use case uh we have 350 to 500 million people on in crypto right now there are 5.5 billion people on the planet and they need real world use cases and crypto can actually solve that problem in a better way that rewards people in open source and and provides a whole bunch of utilities so I'm I'm really excited about real world use cases yep thanks for that um just to quickly introduce myself I'm San and I work at the ethereum foundation leading their Ai and public goods track um kind of like like a use case that we are particularly excited about in the EF is uh having AI answer questions which are verifiable so like one of the use cases we are doing is something called Deep funding where we have 5,000 open- Source repos that make up the ethereum universe and we ask AI to give weights between these 5,000 and we use jury data to see which AI is the best so I think in general we are very excited about the concept of a leaderboard of AIS that we have many agents giving an answer we have humans giving a few of the answers AI can give a million answers humans give 500 answers but we use those 500 answers to select the winning AI which gives the answer for all million so we are applying model for like like deciding funding between repos doing Cil analysis um looking at like predicting misinformation of posts on community notes and on Twitter so those are some of the use cases that we are kind of particularly excited about at the EF um so I I'll I'll move next to kind of the fundraising part we are obviously in a bit of a downturn in the AI market so how have each of you tackled the fundraising question for your projects uh when it comes to Ai and crypto I mean we raised money over a year ago or about a year ago at this point in time and it went really well that's good timing yeah yeah uh for for collab land uh we raised in 2021 uh and just to be honest it's been tough because collab land is a SAS based product uh and so we've been through the ups and downs for all the founders out there uh been in the trenches uh we are now profitable so uh thank you uh it's not been easy we've been doing this for five years uh so uh yeah I don't know it's a great question now uh do we what do we do I think for us now that uh we don't need to rely on investment and funding do we take it to the next level and try and go out there now for funding or do we now figure out how does a sas-based business become decentralized there are not many examples of that but I think that's more of a kind of a interesting harder problem and as like a Founder I like hard problems uh so on the club line side you know that's what we're doing and with mother I think what we're trying to do is try to do something that is fair right I think with the whole meme coins and all of that I I I think that there are better ways of community raising and so we're actively exploring that I just quickly like respond once before moving on to David which is that uh the virtual steam is interestingly enough they're trying to tackle two issues one is how do we disrupt the SAS industry with all of our AI models that's one of the things that they've set their uh eyes on and the second is um how do we get highquality model Builders to come and launch agents on the virtuals platform so interestingly I see some alignment with like you actually like with Mother launching agents and with collab land wanting to disrupt the industry yeah yeah David probably not too much to say here but we recently had a fundraising round um I think if you build something valuable and you know you work hard enough eventually you get your funding I mean crypto is probably one of the easier places to uh raise funds um maybe you know the more spicy take would be that it's sometimes too easy and a lot of low quality as a result uh uh comes about as well so I think as an industry you know if you look at it at a macro perspective I think we attract a lot of capital and if you compare that to Pure AI the output isn't that great so I think we really need to level up uh well so this is a bit of a complicated question for us because uh our our CEO had the foresight uh uh to help invent the industry uh which really kind of helps you get VC meetings uh and you know also had the foresight to create a L1 a couple of years ago that has a large Foundation that can partially fund it but um near AI is actually an equity company not a a token um and uh we so we've been raising from traditional VCS and uh actually to counter uh some of what was just said um we're actually not having as much of a problem uh with the VCS really kind of agreeing with us in in that mindset is it's really the the number one thing that they want uh at at valuations that would make it worth it to not raise a token round is uh is really actual Revenue actual traction and if you're really pulling that like if you're profitable or if you're you know uh growing ridiculously quickly VCS just seem to love that I don't know why yep I think that's great to kind of get such a balanced take from so many different different uh projects that are working here um I think we'll move on next to the fact that AI models are inherently non-deterministic and that kind of causes issues when you're trying to do some of the use cases that you guys have mentioned of like uh abstracting away crypto's complexity so how do you kind of have reliability with non-determinism in AI you want me to start over this time um I really like uh telling agent are the people who are building on our platform uh to leverage L chain tools as much as possible and Factor as much as you can into a deterministic thing with a trigger of uh of the uh you know inference so like you know have it go to an actual logical thing yeah I guess what you're saying is like mix rules based with like generative Ai and that's also the way I see it I mean um you wouldn't want to give a wallet purely to like a generative model with like a loop let's say um you probably want to apply some wh list some transaction guards whatnot that's certainly how we architect the Olas agents um and maybe you know in the future um that will get better uh as the model architectures get slightly different um uh but right now that's the way to go so I I I I would take a step back and take a look at just how the interaction of a transaction happen happens right now everyone is like used to EAS so you have to like first send let's say an agent funds and give it some assets for its wallet I think now with like 4337 7702 I think now you have new interaction models so you can have instead of push base where like you can't give someone temporarily your private key it's not something you can't unknow right so instead of having a push based interaction model you have it pull-based so you can give not the asset to the agent but you can give it a delegation and then there's these there's a whole design space I'm super excited about because now you can have pull based transactions so it can pull F funds from your smart account and with that you can actually have like these after hooks or caveats so if you're saying to an agent it's a swap agent and you're saying I want to swap from base to like arbitrum for example you can say Okay Swap this amount and then the caveat would be no more than like x amount of percentage uh slippage right because the it's not only about the agent not hallucinating but the market conditions outside the agent's control needs to be also kind of enforceable so I think this like pull base model with smart Accounts at the kind of core is a new primitive interaction transaction primitive I think is very much underexplored that I think could Ben agents can benefit from as well the way that I look at it is agents are actually the new distribution Channel and user interface at the same time and so when you're talking about using Lang chain or llama index or even Eliza right like we're seeing all these agent servers pop up we think you should use whichever one you you want we think the important part is to standardize the naming and the the be the ability to network and to negotiate and so if we do that we can actually be in a spot where these agent servers are the distribution Channel for all these composable web 3 tools and so um pulling these together like if we actually just put if we assume that we don't need to have an agent uh that's perfect today what we actually need is an agent that can work with all of our ecosystem now we're in a spot where we can uh standardize demand once demand standardized these agents can scale very quickly so uh we uh we were just side noting on on this one so we at least I I won't speak for you uh kind of disagree with that in in my opinion uh pretty much everything should go through the same interface in between agents uh and uh the agent itself should be doing that forking instead of uh trying to go to like SL payment or slash info or anything like that it should all happen in natural communication uh to me so you saying you're saying all natural language based no Json standardization no I I mean as things continue to get better uh I mean sure you should have a metadata. Json for your agent so you know how to like register in a registry but like uh you shouldn't have like end points uh for indiv any individual thing it should as much as possible as things get better it will all become natural language tossing between everything for sure totally agree I I know the yeah I agree I mean the uh like multi-agent systems agents has been around for a long long time and often in crypto we sort of ignore history and uh like there was all these Asian communication protocols they never really took off because of a pain in the ass to integrate them I I I don't say like standardization isn't useful but I don't think we need like everything super standardized like in know youc 20 in a ESC context where let's say you have like a token standard and nft standard and this thing because that's actually crucial for a uh uh blockchain but for agents the whole benefit is that they can interpret things on the go so I think we'll see this mix um of of and right now the the problem to be solved right is like it's almost like a mandal br set of Agents when you really think about it right and so if these agents need to be able to use services that 10 different groups that are here are using by standardizing it we're standardizing the actual usage of tools and the actual way that they can communicate and pay one another I absolutely agree on on you know some of that part I I think where the real standardization is going to need to happen is inside of the agent Registries itself and uh there will likely be if any of you have used artifactory or something professionally you know there's a a local corporate version of all your agents and then there's a forward uh Cloud thing or like open source hosted whatever thingy uh and uh when when you when you think about it that way you're like okay all these things are going to have to have different Discovery systems inside of them they're all going to have to know how to Fork probably as a natur natural language prompt into those Discovery systems and then decide to just run the sub agent instead of having to like strictly Define a swarm it these these systems will naturally progress over time and have fallbacks like an operator or like some other uh so so that's kind of where I see that going so apis go away uh actually uh we just launched it this week too you uh give API inpoint uh and we create full agents in seconds uh not API endpoints API documentation just you could just yeah a apis are going away for the most part uh they will still exist because you need a thing to communicate with it but it's not how we're going to do it but apis are for humans I think it's like python versus strictly typed languages right it's like natural language allows us to be fuzzier and we now gotten that I don't think we'll go back to strictly typed agent everything like one standard one registry as long as you know that some registry exists the can find it read it interpret it go along and and the way we solve that is by having orchestrator agents that actually you can say like this orchestrator agent is the one that's going to pull together uh the set that's going to be used for this task and we don't uh we don't assume that we're going to be that orchestrator so we designed the pattern as opposed to providing the service yep I think it's great to have a panel where where everyone is disagreeing so that was like this was a good panel thanks for starting that Jay appreciate that um I think we are at time so maybe just final thoughts for everyone if anyone wants to talk something yeah I'll go last you goe every single app is going to be turned into a series of agents and I believe that web 3 can actually form the capital bring in the communities and actually create competitive agents in a place where they're used to web two ways of doing apps yeah I think uh this year you're going to see a bunch of different agents like millions of Agents but a lot of them are going to be be useless so yeah I love this idea of the web of agents and I I subscribed to it and I think it's also very important for us in crypto to enable real ownership of agent like sovereignity over your agent and kind of being able to control it um have it be personal to you um I think that's something we can really add to the broader AI agent space so um I'm going to kind of cheat here because I have a final thought uh and I really want to say it however it's getting announced uh at uh during Ila's talk at 3:55 on the regenerator stage at the tail end of it uh some of the people on the stage are are participating in that it it's going to be a thing so if you're not booked at 355 to uh 425 give or take I would hit up that talk especially for the end it's I'm excited I'm really excited you can give us a hint at least um crypto okay I I I'll say what I want to say which is crypto is a a balkanized factioned kind of a a thing we we get on crypto Twitter we fight against each other we say your coin is uh what this place smells like uh that you know um and we shouldn't do that especially within the AI space we should realize that our biggest competitors are going to be you know uh Apple anthropic open Ai and when you realize that and you realize you're competing against trillions of dollars what we're fighting on out over here isn't much at all we should consider working together being open doing the same thing we as a lot of us got into crypto for in the first place which is not token go up it's changing the way the world works and making it a better place so maybe that's kind of the announcement y That's great yeah thanks a lot for this panel yep