Agentic AI Panel by Harpreet Singh Maan

Transcript

Next up is the much awaited AI panel. We're diving into the world of agentic AI with a panel of pioneers moderated by Danielle Tickner, CEO and founder of Wource. This discussion features Harprit from IKA exploring intelligent knowledge agents. Justin, VP of gaming virtual protocol, Vader, founder of Vader AI, delving into AIdriven virtual environments, and Artac from Chain GPT, integrating AI with blockchain for smarter contract interactions. Together, they will unpack how autonomous agents are reshaping the decentralized landscape. Get ready for a powerful conversation. Let's bring them on stage

with a huge round of applause. All right, thank you everyone for coming back and we apologize for a little delay but happily we're all here right now. So the panel is dedicated to Agentic AI as Erin just announced and um I want to share that um in February we had uh this discussion with Harpit. We were also expecting virtuals but you guys couldn't make it.

We had the discussion in Hong Kong in another event and we were talking about agents like uh what one agent could do and how how we define agentic uh AI and all kind of other questions and in the end of the panel of the panelists and Harpit was one of them told me uh Danielle it's not about an agent anymore you know it has evolved we're talking about multitude of agents so I promise that the next panel this is what we're going to discuss and the question goes to you Harpit. Can you cover a little bit their transition from working on one agent to working on multitude on an ecosystem of decentralized agents? Uh the mic is right

there. So we have two mics here to share. Hi. Uh first of all uh thank you. Thank you for having us on this panel and you know like and uh Bri bringing back the discussion right that we had in Hong Kong previously. So getting back into uh the field of agents and agentic itself today you know like we are like looking at this shiny new ecosystem that is emerging but in research it's been there for a very long time where we've been talking about where intelligent agent ecosystems or agentic ecosystems are able to solve things on our behalf or do things on our behalf where we are able to delegate to computer systems to

solve our daily challenges like today you know like people are running out of attention attention span. We have so many apps to deal with, so many works to deal with, so many you know like computer systems to manage. Can't we just like have something to help us automate things, some agents to help us with our daily task, you know, like once we move on that phase, the next phase will be like can we work rely on robotics to you know like even simplify other parts of our life. So that was like a lot of the elements in research. So every field right we have to test it on the you know like the less risky approach first like today asking a bunch of questions to an uh LLM doesn't hurt anybody so it's simple right people can try it's low hanging fruit you know like

you can you know like uh execute those side of task but asking an agent to move your finance or managing your finance and your portfolio is a much riskier positioning right so when you talk about D5 AI or you talk about decentralized finance agentic finance per per There's a lot more challenges there. But at the same time when you look into uh in in a in perspective how even humanization evolved over the years right looking into when we were doing things on our own when we were trying to survive on our own in the in the forest for example and then we tried to become in the hunter gatherer approach and then we went into civilization and then we went into globalization. How much have we progressed right where we relied on

other people's expertise, talent, resources, growth to actually push our industries or our economy or our civilization further as well and a global civilization as a whole. So imagine the same concept. What can an individual agent do? What are the individual agents uh benefits and strengths? But when it goes in multi- aents, it becomes even bigger. Now every agent is able to do something specific, right? Let's say there's a uh expert DeFi agent defi DeFi AI agent but what say there is even more specific let's say one is specialized just in Bitcoin financing maybe one is specialized just in Ethereum tokens right so you can break it down in many categories for each of the domains and it exponentially

grows the you know like the whole ecosystems the intelligence in the ecosystem the automations in the ecosystem and I think we are just in the very beginning part of it. There will come a day where we sit down here and you know like an agent is managing a big part of our life and we can delegate to it and wake up on a morning where we don't have to worry about getting our coffee or breakfast or you know like getting your car to your work. It's all done right. You just execute on things that creates best value and and and focus on maybe doing something good. Yeah, I like it. Of course, there are all kinds of risks that come attached but this is a separate subject. So um the next thing that I wanted to ask uh you

guys and this is question to both of you Vader and Justin um you know when we talk about autonomy of agents or autonomy of a genetic AI the another dimension is predictability. So how do you manage that today in a in a in applications that you guys have developed? Where do you put the stop where you see that the agent is doing something wrong? How do you uh basically rectify that? Sure. So we put in um I'll give you example. We used to run an investment agent and with an investment agent things can go wrong very quickly. Number one the agent can make a huge trade that

is terrible and can make us broke immediately. So um I mean simple we just put in some guard rails like if formless uh to to prevent that. But I think if we're considering more like LMS that are speaking uh that are chatting uh you don't want them to say some unethical stuff or you don't want them to talk about certain stuff or like u you know some some LMS that are developed in certain countries might also even get political that they don't talk about certain stuff. A bit of a censorship it sounds. Yeah. Yeah. Um, yeah. I mean, us builders, we usually put in the

guardrails to prevent them. And I think though, like what's a bigger problem is um so number one is you don't want the agent to do specific things. You know, you can write negative prompts and you can you can prevent some stuff and but the more rules you put in, the less independent and creative the agent becomes. So there there's a trade-off between um how do I say like hestinations and fine-tuning. Um on the other hand I do think it's really important to provide labelled data meaning if you want to find if you want the agent to do

certain thing well and if you want the agent to do it in this way rather than that way like you would need to train the agent with that data and fin that yeah anything you want to add thank you yeah just I think for me I'd uh like to add on the protocol end. I think that as a builder like Vader, he's more expert in making sure that the the agent doesn't break. Uh I'm not the CTO of virtuals. I'm actually the VP of gaming. So I think uh if you have guys seen the phases of virtuals, we came from a app application meaning that anyone can just launch an agent. There was no bonding curve in the version one. Then later on we had version two where there's bonding curve and VA is the first one who graduated

from the bonding curve if I recall correctly. Correct. Yeah. And then third, we have launched our Genesis version whereby uh instead of a bonding curve, uh you can just insert uh virtual tokens and points and if it doesn't hit the contribution required, you'll be refunded the same amount as you've inserted. So there's no PVP before the uh before the launch. So it creates a more fair uh user experience for everyone. And yeah, I think that's something that we are gut railing on our side to ensure that you know it's not a a DGEN thing. I mean event is all about DGEN but we try to protect our users as much as possible and we did compensate uh certain exploits as well because we understand that that uh user experience

for everyone is the utmost uh important for us. Thank you. So question for you art um more about infrastructure. So I know that uh change has announced that you guys built in some AI infrastructures and you're going to be focusing a lot on AI. So can you share what you guys have been doing lately? Uh sure thanks for asking. So basically uh we've built an infrastructure we built an uh web three native LLM model that is uh doing not general search or general uh languages about like everything around but it's more concentrated focused on the web three

stuff on the stuff and so on and using this LLM model we created different um SDKs APIs for uh for any AI company or AI agents that can work with. Uh so currently we even provide grants for it and people can use it to create their own um NFT builders with it. So like it's a picture generation with bu putting it in the right smart contract. We have uh SDKs for smart contract audit. So basically AI agents that is doing just smart contract audits it it will make

the audit like security part of the company much more uh cheaper that it should be. So um different kind of stuff we have we have a chatbot that is also uh working that so basically company can integrate it into um into their project and users would be able to ask questions about the product and there would be answers from the crypto perspective. So that's basically what so you're basically building a suite like an AI suite for people to onboard. You're right. I I even will say it's more like an infrastructure part

for any other project that is building in AI stuff uh web 3 AI to utilize it and uh get their users to work with it. Got it. Thank you. So um I was looking for uh someone who could give an input on uh AI launchpads and I understood that better you can help us with that. Yeah sure sure AI launchpads sure where do we stand? I think Justin also gave a great brief. Uh the biggest problem right now with launchpads in general, not AI or agents is uh so right now we have fair launches which are popular, right? Something like a pump fund. You go out there, you put you launch a

token, you buy some of that token yourself, and then the token is out there. And what usually happens is that the moment the token is launched, snipers snipe it out, buy it before anyone else can buy it. Now, if this is a random token that you're not going to continue, there's no problem. It's it's a loss of snipers. But if it's a real project that there's a team behind it and people want to build something long term, um it is a problem. And the main problem here is let's say um Project X, right?

Project X, the fair value of Project X, let's say in all of us, our consensus is we think that this is worth, you know, $500,000 market cap. And if it's below that number, we're happy to invest. If it's above, you know, we think about it at that point. um you know let's say a a token is launching and we are not able to get in at 500,000 market cap we get in at 2 million people who get in at 500,000 or even below that are snipers and snipers just get in and then flip like they don't add any value to neither projects investors anyone so I think um

so virtual has a new launchpad design which solves this problem in a way that anyone if they have points which is like more detailed anyone can invest at $120,000 120k market cap and the token is launched at 360,000 market cap. So immediately you're in the money like you start with 3x. So, it doesn't make sense for a sniper to get in and snipe. And even if they do, the pre-sale investors will always be getting in at an earlier valuation than the snipers. Uh so, happy to get into more details, but just wanted to

say that I think that's a huge revolution. Number one, and number two, you know, a lot of launchpads are just uh for memecoins and rock projects. And that's a huge problem. If you know, I always see crypto as an alternative funding option. If you don't want to go to VCs and you want to you crowdsource whatever you're building, it's a great option. However, like pumpf fun is where the liquidity is at right now and you cannot raise there as a legit team. You need to come up with a memecoin or something that is trendy, something that's not going to live long.

So I do think we need more launchpads that embrace utility projects. Cannot agree more. Thank you. Uh going back to some of the technical aspects Harpit with you. Um there is a bit of a issue with data formatting and standards when agents are using all kind of different platforms and infrastructures. Any inputs on that? I think going towards the future right I think the next programming language is English right today you can see Chinese yeah I mean even for example that's where in the western world in the

western sphere is where a lot of the innovation is happening language will become the programming language if you want to put it together where today uh coding you know like it's getting where you're wide coding and you like coding lesser and less it's more you just need to have the knowledge to understand that it's doing the right thing and this is how it should perform. Execute the right thing, you execute it, right? At the same time, when you talk about communication between uh agents, right? Previously, we relied on a architecture where you had to API, we had to have like specific standards and we started going towards open standards. Uh and now except you're dealing with legacy systems, right? you don't require that when you're dealing with I think the communication is going to move towards where systems websites applications

servers will have their own agentic interface right where you can have an agent who manages the data protects the data because it gives it more autonomy right when you call for an API can be anybody calling it but when you have an agent there it can actually check right is this person malicious is he doing something wrong right it also add a other level of security also complexity I will say a bit of delay maybe but at the same time it improves communications uh in the outside world in terms of communication standards I think we're going to see a shift right in terms between how agentic and agents are going to communicate compared to how traditionally we've been coding applications for server applications or traditional computer applications and

not only that we are at a point of big user behavior change as well right today we are like glued to our phones. We have to keep seeing notifications after notification replying typing right? Communication will be the future where we actually can just hear out somebody can type the text on our behalf or maybe communicate on our behalf and execute task on our behalf. So today the interface is still a bit boring. We still have to go type it down into a chat prompt. But as that improve, as uh speech to text recognition improves, it gets better, right? So you know the future movies that we keep seeing that people are just wearing headset and doing all their work in their life. I think it's not very far away from now uh

and all the technologies that are needed to build that future is getting ready. Yeah. Thank you. That's a promising future. So uh we have covered quite a few subjects. We need to look into gaming as well. So I was going to ask you just on the word is that you guys launch any game very soon and how the game is going to look like when the game is made by agents. Okay. Uh what we're experimenting now is what we call modular NPC modular AI agent for NPCs. So what we mean by module is that we can plug this agent into like Roblox but at the same time we can also plug it into like sandbox. So one agent that we're experimenting is a hideandsek. I'll put it in simple term is that these agents

are given like 10 different locations and the reward they can select from the libraries that are given by the developers as well. So the agent will also learn like is it like hard for people to find me. If it is then he stick to the same location first but it gets easier. He might move one location away in a random out of 10 because able to read the map right determine the coordinates like a very unused coordinates or stuff like that. So yeah that's one uh very simple modular agent that we're doing. We also have like uh we're building what we call basically a pad that follows you around because our framework the virtual GA framework is able to perceive the environment. It can know that whether you are you're dying often like what

weapon are you using it can recommend you an item. So that's that what that's what we expect to drive the revenue up for the developers and all the other agents are also complementing to drive up the AU all the gaming metrics that we can uh think of. So that one will be parked under ready gamer. So which is a JV between virtuals and also software run. Yep. So this is the multitude of agents in action in gaming right very good. Thank you. So uh we have covered uh data we have covered from agent to multit agents covered launchpads all kind of gaming. Now there's one subject that many people would like to hear about and that's for you art and this is a very short input on you are looking a

lot into investments. Right. Right. So uh could you give us insights and what are you investing in right now with everything you have heard on the panel and everything that you see every day with DEX and so um I'm currently mostly looking at the projects that simplify uh human work with DeFi. So it's called usually DF AI. Um I really want to find the working product that would be so for as for now what do we need to do to make like a bridge from Ethereum to BNB chain or anything else like we need to go to B we

need to buy like the native token of the chain we need to like go to some DeFi app and like everything like that is very very complicated and will take some time even to do so or use a centralized exchange, right? It's it's also quite an easy I'm happy that I'm not only the only one challenged with that. Thank you. Yeah. Yeah. And uh we see quite a few projects building already such kind of solution like Griffin AI and uh and others said you can just type like I want to swap my my Ethereum from this chain to another and stake.

and it will find the best uh the most like cheapest way to to bridge it uh the DeFi application with the most high yields generated and will do everything by yourself that you just need to to tell what you want to do. So, and uh recently like I remember Alsar Ventures I think they posted about the like narrative of post web. It's like the new after James favorite. Yeah, he's promoted it. Yeah. Yeah. is promoting a lot and I love it like because in the I I see that soon we will be able to work like agents will be able to do

everything in a internet for us and I'm looking at that kind of agents AI agents uh and that that would make our life easier makes sense thank you for sharing so to wrap it up just one last BL blitz question like really short answers if I can ask I can I mean I think we can do it really really quick. So uh if everyone could say like just a short sentence of what excites you right now. What is the most exciting thing in AI for you? I think for me because I'm building the AIX uh gaming intersection.

I believe that AI gaming will be the might be the next meta in web three as well. Right. Currently we are building more on the web two side because there's a lot of active users. you can experiment there more but I'm sure later on there might be who knows uh gaming agent launchpad or something like that that we're trying to build in the future. Yeah. Thank you. For me personally I think LMS are getting better and better. So it's no agenda just personally I I love the I'm playing with them and they're able to do much more much much more. And I think soon we're going to be able to integrate them into our like for example um go outside and walk walk around not during the day but in the evening in Dubai and

you know ask GPT okay GPT where should I go for dinner you know where should I walk it can't tell you or it will tell you something but it asks your location first because it's not even connected to Google Maps or Apple Maps. Um so it gives you a very like blind information even though we already have a lot of tech possible. We just don't have the integrations. So that's what I'm personally excited about. I recommend to go to Dubai Mall. Best air conditioning. Thank you. Go ahead Harp. Yeah. Uh I believe uh a decentralized intentriven multi- aent system is the future, right?

So basically it has to be decentralized because we do not want it to be fully centrally controlled and your data be handed over to you know like some centralized regime right controlling all the data. It has to be user intent driven. So in this case based on what user wants to execute where you're able to delegate your actions and activities right multi- aents because a singular agent can only do much. When you put it into an agentic economy it grows. it is able to do a lot more execution and it also will require its own financial economical principle. That's the reason I actually the one of the main reasons why I got into blockchain because I knew that this might be the best way of implementing the financial models and securing it behind the age economy. So

those are very forwardlook looking futuristic things right that it might sounds alien today but it'll happen in the next few years because as computing power gets better as LLMs get better as you know like all the agentic AI models get better camera technology gets better all of them are like kind of exponentially pushing this reality to come from so with that thank you I love the intent part so art for you oh I will say I agree with All all of the panel all of them are amazing. Yeah. And I love AI agents in NPCs. Like I love gaming and I hope I will we will see the the time where like we're talking with like NPCs and it would be crazy. So uh

but how I said like mostly I'm more focused on the way of uh defi. So making this kind of defy life easier for for people and And like if we are putting web three away of course the AI agents that are providing opportunity to develop like cursor is also something like it's a no-brainer like who will imagine 10 years ago that it would be possible like one people currently can create one billion dollar companies already just one people one founder. So that's crazy. It's true. The innovation

has been impressive in this field. Very very accelerated. So thank you all so very much for coming today and for sharing your insights looking from completely different angles. I hope that it was very uh insightful and useful for the audience. Thank you all for coming and listening to our talk. Thanks. [Applause]