How Fetch.ai is building autonomous agents for the agentic economy

0:01I'm Ashen Addison from the Cryptocoin Show and today on blockchain interviews we have Humayan Shake, CEO and co-founder of Fetch AI. Here to talk about decentralized AI, autonomous agents and the open intelligent machine economy that is coming. Uh Hayan, thank you so much for taking the time. Appreciate it. Thank you. It's a pleasure to be here. Likewise. Excited

0:25to dive into uh your expertise in AI. over the past few years, it's exploding and it's hard for people to catch up. Even people that are in crypto following the news every day. AI is moving so quickly and I know you've been studying it for a long time. I'd love to start out with maybe a little bit on your background in studying AI and and how you got involved with Fetch and then we

0:49can talk about all things AI. Yeah, that's great. Um so my background computer science uh engineering and uh got into gaming quite early on probably uh you know that was my first kind of love uh into gaming building games that's how I met uh Disr deep deep mind founder and we got connected 200 probably five 2006 then we

1:21kind of started discussing I think 20089 we set up Deep Mind which exited to Google in 2013 and I was looking after the commercialization of AI which means you know how are we going to take this technology how are we going to go to the market with it and how are we going to make it into revenue generating kind of technology looking at it the technology

1:46itself and then finding ways to adapt to the markets. So that's that's then 2013. Then I when we came out of Google it was quite clear commercialization of that will not be that easy uh unless you're a big company a big enterprise because you have all the data you have all the all the relevant tools. So my focus then became how can you bring it to the masses and

2:13what uh real component would enable you to do that and it became quite clear it will it will be like agentic systems because agents you can bring agents uh individuals can interact with agents you can break out the tasks use agents to fulfill them and that's how the whole thing will kind of evolve and on the other side it was the language which was going to be the

2:38because the interface was going to be language and the the ability to execute will be the agents. So that's how I got into the agentic system. So 2018 2019 we started up fetch uh with that premise. So we've been around a while um as as you probably know and uh people used to think when we talk about agents it was crazy not going to happen. It's just a

3:04narrative which we've created. But I think uh we this they they stand corrected on this one. The the people who were the naysayers um we've always tried to stay ahead of the tech um and you know kind of create the pathway. Um so if you think about agents we were the first project deep was the first company which kicked off this whole AI phase. Uh we then did our first merger in crypto,

3:32the biggest one we could do. You know, there's pluses and minuses, but time will tell. And then uh yeah, we kind of we're trying to build, you know, ahead of people and something which brings utility to decentralized space. Uh really that's kind of an overall journey and here we and here we are. And yeah, I'd love to dive into that merger in a little bit about ASI and you know what's

3:59the reasoning behind that and and why it's so valuable. I I want to ask one more question about agentic AI and you know since 2018 with the uprising of LLMs a lot of you know regular everyday people have started to use uh language to communicate with with agents per se but I feel like they people still don't understand the full functionality and capacity of what it is to become because

4:25I feel like there's a lot of functions that we still and we still we have to ask AI and we have to do it ourselves for for the most part. Um, and I feel like in the coming years all of that will be taken care of and and it's going to be so much bigger than it still is now. Yes, I think we we're just at the start. Well, it's not even the start. It's, you know, just before the start.

4:49Um, but we we're on the journey. So, so what what I mean you I'm sure you have talked about agents so many times. Do you actually have an agent? Did you create one? and how yeah yeah I think there's a lot of people that you know there's a lot of discussion around AI agents I think for the most part people are relying on these large language models um and I've personally used a few

5:14AI agents here and there and telegram things like that but even being involved in crypto I don't think I'm taking advantage of it as much as I could be and maybe the barriers to entry and understanding what are the things possible and how do I actually get started doing that to help me run my business or my everyday life. I think a lot of people it needs to be broken down

5:36a little bit more still. Yeah. So, so, so that's the kind of tools we're bringing which is how can a person who is not a coder who doesn't want to be too involved in technology can can engage with technology. I think chat GPT did a great job where it gave gave this LLM models in the hand of everybody and they started using it and they started using it because they didn't need any

6:03technical help. Right? So that's what's going to happen with agents as well. Uh you will have your own agent. You will be represented by your agent. You'll have two lives effectively. One is the professional life, one is the personal life. And then the agent will uh take uh input from you or your intent, your your your command, your demand, your you know all

6:28of those things and it will interpret that into actions and then carry out those actions. So the first step in commercialization where people can actually start using it is provide something very simple just like charge GPT you should be able to set up your agent and then you should be able to communicate with your agent no different to what you do with chat GPT because

6:50that's just language and then you can ask it to do things and then it goes and interacts with other agents to actually carry out you know it could be just the textual communication it could be more execution ution, it could be generation of something, it could be any of those things and that's when you start seeing the traction. Um, so so my my belief is

7:15anybody who has a website or a social media page or a professional page like LinkedIn page, all of that will become attached to an agent. So all of that will become agentic. And if you think about just that, it's a huge huge market. So you actually see what's coming. So all of those will be agents and you know you could have every company will have

7:43an agent or multiple agents. So some will be internal, some would be external, they will be represented by agents and individuals are the same. But but but let's say if if I play two parallel scenarios, if you think about what happened with web, everybody went off and started creating a website was very difficult in the beginning and became very easy now. So you can deploy

8:07it, click it, click a few clicks and you deploy it. That's going to happen to the agents. Each website will become agentic. So it's where you have a website, it will be an agent. So and the reason why that would be the case is you would have you know the websites give you information but it's one way you go you interact you you do the actions agents will be more interactive more

8:33intelligent more communicative you can have different things they can do so it will become more dynamic it'll become alive so that's that's kind of the parallel you can imagine then comes the e-commerce which was on the websites we built e-commerce solutions The same will happen with the agents because they will need to interact. They will need to transact. So that's that's the journey

8:56we've just started and all of that is needs to kind of not be fitted back into web two. It needs to be AI first now. It's not web first. It's AI first. So what what does that mean? What it means is that you shouldn't be doing the work. You need the large language models or AI to actually do the sum of the work and then bring the solutions to you. And

9:18that that will change the game quite a lot o over the next few years. And that's how we're going to start getting using AI a lot more than what we do today. So we we're not there yet. We're just starting the journey. Definitely. And what needs to be done in do you teach the the agents, you know, how to take those actions based on the the the command and like how exactly do we get

9:45to that point? Yeah. I I don't think you're going to need to teach them as such. It's you carry out the actions, they will carry out the actions for you and that will result in the learning will happen in the background. So you don't really I I don't feel we need we will need to do anything. I think it'll start from small tasks. You'll guide the agents to do

10:10those tasks and then suddenly they will start learning it. They will start doing it. Of course in parallel you're building the technologies you're building new models which are much better they know what they're doing. So for example fetch has a a model which is very tuned for agentic systems. What that means is when you go to chat GBT you say hey you know do this for me it

10:33goes and does that um textual form graphic digital form and then but but when you say to it okay you know now connect me to this person's agent you can't you can't connect to agents right it's all one model encompassing which goes out to do different tools but these tools are dumb for now and if you think of them to be more alive and dynamic and intelligent they will be the agents. So

10:58what we have for example today is and you can you can actually your listeners can go and test it out on ASI1 we have what we call is an agentic LLM. What you can say is hey I need to do something and it will go and find the agents that can actually do it. So people outside of the LLM can build the agents and deploy them and you can actually find them by

11:24going onto a normal textual conversational interface. Mhm. That's really interesting. I'd love to know more about the ASI, you know, I believe it's artificial super intelligence merger and what was the, you know, the reasoning and the value behind making this one big unit. Yeah. So, so if you look at what's happening in the world, you you don't

11:52just you don't just build models because you need compute to build those models. You need data to build those models. Then you also need to commercialize these models otherwise what are we building them for? They need to do something and that's the application layer. So if you if you work out the this AI stack these the stack starts from right on the top there is an

12:16application layer. How do you build these applications? If that's agents, you come underneath that these agents need tools and needs integrations and these agents need learning methodologies, algorithms. So all so that's that's your uh machine learning models that need to to be built. Then you come down further to do all of that you need really data because without

12:44data you can't. So you need data tools and then if you go further down these models need to be deployed somewhere they need to build they need to have compute. So you need a compute network as as you saw I mean Nvidia is shot up became one of the most valuable companies in the world because everybody understands this AI phase is going to need a lot of

13:08compute. So you do need the compute. So owning the whole tech stack or actually having an integration between the tech stack makes it a lot more efficient. And if you look at especially the crypto part of the world, there is no project which delivers that whole tech stack. Mhm. There is no project where you can go right from the I'm going to build the

13:33agent. I'm going to ask the agent to deploy a machine learning model on this compute and I'll run it and I will make my agent available and people will be able to interact with it does not exist. It's the whole tech stack and then on top of that you need a lot of research to keep building these solutions for the next generation of AI. you need to stay

13:53on top of the so taking that all those projects which includes fetch AI singularity ocean and kudos putting them together in a way that the tech stack makes sense and is much more efficient and is much more um encompassing as compared to anybody else and uses one token ultimately to use all of those services rather than going one

14:21token convert to another token, another chain, this and that, the other. You know, that's if we want to build a solution which world is going to use, we need to do it efficiently. We need to do it quickly and we need to do it in a way that people use it. Spring the utility. So that was the reason for the merger. Yeah. No, it's really exciting and it's

14:42uh it's you know history in in in the small industry that that crypto is small market capitalization compared to the world that um they always say you know these blockchain companies they're on the same we're on the same team you know it's such a small industry we need to work together to grow into more of the mainstream so I'm glad to see that you know it's not competition it's

15:05cooperation so it's really exciting yeah that's that's really the the the premise of the whole thing and that's what we did that for and actually you know the more the more we are working together the more we see that the tech stack makes sense. Yeah, that's great. I would love to talk a little bit more about decentralized AI and maybe the blockchain part of it, but there's been

15:30this uh this battle through calling out, you know, open AI that it's not really open. It's more centralized AI and decentralized AI has other advantages. Um can you talk about the importance if there is of having actual decentralized AI versus centralized It's a it's a misnomer because I think people use it just to launch a new

16:00project mostly. Um it's decentralized AI. What what is decentralized AI? We need to actually define it. We need to actually understand that what components are decentralized. So so if I if I kind of break it down into what what does it mean to have decentralized AI? AI is not one big thing, right? In in if you if you think of AI, AI is made up of

16:28models. AI is made up of smaller models, large models, large language models, large other types of models, specialist models. They then interact with each other. How do they interact with each other? I mean, they could be agentic systems. They could be just integrations with APIs so that they can actually interact with each other. So, so when I when I think of decentralized AI, what I

16:52think is that if somebody went and pressed the button, can they turn it off? So, well, so if I take that theme, then now we have so many LLMs, right? So, if you don't have chat GBT, you'll have something else. We have our own. How were they trained? They were trained on a decentralized compute network. So if one company says we don't want you to train this model and we're going to turn

17:20you off, it doesn't matter because this is decentralized and no one person can turn it off. Now when we serving inference and somebody doesn't like the inference, can we stop the inference being served? No. Because it's being served via a decentralized system. Did we train it on decentralized? Yes, we did. Do we do we serve it via decentralized? Is it hosted on

17:45decentralized compute? That that in itself makes it a little bit more decentralized than anything else. Now, and you can train it by using and interacting with multiple participants and train it with multiple inclusive data from different participants. That's also adding to that. Now, the biggest part portion is when you go out in the market and do something with it, is that

18:12decentralized? And what that means is you can have your own agent, you can be running your own AI models. We can do all of that. So there is value, but the value comes at a price. That decentralized value. The price is the cost, the efficiency. It's not as efficient. It's costly. Uh it's not easy to do. So all of those things and all of those components need you really need to

18:37have a conviction that decentralized AI is what you want. But most people don't care. Most people care about utility. So if you don't provide them utility and you just keep saying, "Oh, it's decentralized." Nobody cares, right? Only crypto community cares. And that's a small community, right? So what we need to do and and actually even within the crypto community only a very small

19:02are actual uh you know actual people who do care and the rest of them are actually doing it as a lip service because that's our unique selling point. We are decentralized. No that's not that should not be the the the central selling point. That should not be the unique selling point. We need to make utility out of it first. People should use it. then it's decentralized then it

19:27makes sense. So I I have um kind of mixed opinions on it. I think we we need to have decentralized AI and when AGI comes you want to have all those components and we're proponents of that. I'm very strong proponent of it. But what I am not um going to support is a u you know is the argument that everything is okay if it's decentralized. It's not. If it's not effective, if it's not

19:53useful, then it's not okay. Yeah, that's a great perspective. And with businesses that are, you know, there's so many of these huge corporations that have a lot of proprietary information for them building their own agents that, you know, they can put their insider corporate information in there to be more efficient without having to share it with OpenAI, for example, or other

20:18corporations. Is that does that require decentralized AI or just private AI? You you're always going to have a mix of it. Um you're going to have private AI, you're going to have mix always going to have mix mixture because people don't want to share data. Um people don't want to share data. I mean that's that's a that's a point we have all seen now over

20:42the last how many years. data is not going to be shared by people and the only thing you can do is you can train AI model in a decentralized way where people can participate in training of the model but even that is very difficult for people to convince to do because you know if I am spending money and I want to make something out of it I'm not just doing it for sake of

21:05calling it kind of decentralized or not centralized or whatever I'm doing it for actually generating value, economic value, and and that's going to Yeah. So, I I think there's good and bad components of both sides which we'll see kind of playing out over the next few years. Mhm. I actually read that some of the industries, you know, not finance, not cryptoreated, are are outpacing like

21:33Silicon Valley in in AI adoption like healthcare and these other industries. I feel like AI is going to tap into all of them. Have have you seen this? A AI is going to be part of everything. So that's a given. But but you know what is quite interesting is like um I don't know if you know about the the payment uh uh payment app called Pisa which which became very popular in Africa. The

22:02reason was that there was no existing infrastructure. or places or or sectors or or vertical sectors which do not have too much new technology already. they can easily go AI first and they can start building solutions with AI first and that's where the agentic systems come really well because if even if you have legacy systems you can very easily

22:31put an agent on it and you can join this new AI race because it can suddenly bring you into it's a retrofitting mechanism right as much as agents are AI based agents can be very useful even if you don't retrofit them on um legacy systems but they actually they act as a layer between the legacy system and the new AI paradigm. So you can use the agentic systems to bring older vertical

23:01sectors into that. So that's one interesting thing that's happening. The other interesting thing is happening is that research and in science and technology and medical it's it's accelerating at a pace which we have never seen before and that's due to AI. What you have is you can do I mean we you know alpha fold as you all know um it it's enabled people to do what they

23:31would have done in 10 years it has enabled them to do it in one day. So would have so the the pace of research in biotech in physics in all the basic sciences and bringing out new molecules new materials that's going to become very very accelerated and I think that's where we seeing a lot of traction as well. So that's you know so medical for sure biomedical because we all want to

24:01live longer and we want to find solutions to it and AI helps there but you know we have to be clear these are different types of AIs they're not just one big AI which I think is what people think AI is one and there's just everything is in it one large language model can solve everybody's problems. Mhm. Yeah. No, it's a great point. And recently we saw, you know, Trump had all

24:27the the tech CEOs there and they were announcing billions and billions of of dollars of investment for, you know, medicalbased AI and like curing cancer, all these diseases and, you know, immortality, who knows, uh, with AI. So, it'll be interesting to see outside of finance and blockchain that that I'm sort of stuck in uh the the the medical side and how that's going to advance

24:51over the just the coming years. Yeah. I mean, that's that's I to me that's very exciting because that's where you're going to see a lot of traction. That's where you're going to see a lot of people doing um interesting research, interesting stuff. And that's yeah that's that's a space to watch for sure. Definitely. I want to touch briefly on you know the crypto and

25:15blockchain side. You mentioned earlier about you know the agents needing to transact on your behalf. Uh is there infrastructure with fetch right now for them to have a wallet and pay with stable coins and you know interact with different chains? How does that work? All of that is all of that is in place. Everything is actually we're not talking about it. We're not writing papers about

25:40it. We are actually showing it. It's available right now. So when you go to um our own large language model interface which is called ASI1.AI, you can actually interact with these agents right now. You can build your own agent within a couple of clicks. You can actually deploy it and you can actually speak to it. Other people can communicate with it. You can communicate

26:07with other agents. So for example, agents representing machine learning, other machine learning models. They could be representing any any of the hugging face models. You can actually ask it to do protein analysis. You can ask it to do medical uh analysis on a scan. You can ask it to generate a report. You can do all of those things using agents. Or you can say, "Hey,

26:28connect me to Ashton." I could say, "Connect me to Ashton." If if you have an agent, it'll find it. it'll connect me to you and you know our agents can do whatever they need to do. Um but the value exchange the economic value exchange is very important. So all these agents have their wallets they can make payments to each other. They can actually have payments in escrow. We're

26:50building trust mechanisms so that people people can trust who the agent they're communicating with is it real or not real. So for example, if if you have a website, then the agent also gets the trust level from the website. If people believe in your website, they will also believe in your agent. So that's kind of all the trust mechanism that they they're not they're not the future, they

27:12are the now. So it's happening. It's very cool. And with the you mentioned like building some trust mechanisms. I'm sure there's so much more to be built. What's part of the focuses for the next 12 months in growing ASI one even more? Well, our focus really is to get this technology in the hands of everybody. Everybody doesn't have to focused

27:39enterprise or anything. We want this as I was saying when you go in a gathering of AI people uh you say hey I mean you've heard of agentic systems they yes but do you have an agent? Then 95% of the people actually in some cases 99% of the people have never built one or owned one and that's got to change because you have to give something to just do it

28:04very easily very quickly and they have an agent they can interact with it. So that's step one. So my biggest focus at the moment is commercial in in terms of commercialization is what Chad GPT did which is let's put the technology in the hands of people and let them use it. tell us what you want from it and then we will build that and that but because

28:27because the infrastructure is ready and of course it's going to keep improving and keep evolving and keep changing but the initial infrastructure is ready to deliver that to people which means that they can they can actually take an agent they can interact with the agent they can ask to do some tasks and then then the whole evolution starts How how do you use the agent? How do you use

28:52the AI behind the agent? What do you do with it? Uh you we know we know LLMs are good for completing um your your modules, your academic modules, you you know writing papers and doing the textual things. What more can we achieve because there's only so much you can do there. So that's um that's our push this next 12 months. We want to onboard businesses. We want to onboard

29:15developers who can build machine learning models. They should all have agents being representing them and people should be using all of that technology just from a simple interface. It's very exciting, Hayan. I'm going to check it out for sure. I'm going to start today uh ASI1.ai. I'm going to I'll put the links in the show notes as well. And I think it's

29:39sort of the same for Bitcoin. you know, when people they've heard of it and but and they're on the fence but they're like there's something stopping them like why don't you just you know get two bucks you know like just try it and that's the only and once you get it you understand the value that it has and we need to have that for AI agents as well.

29:57Yeah, absolutely. But it's coming definitely and and I know Fetch and ASI are doing a great part in that. Uh I love what you guys are working on. Please keep up the great work. Uh I would love to have you guys back on in the near future to hear about more updates in how Agentic AI and ASI is growing. And thank you so much for the time today. Appreciate it. Thank you

30:19Ashton. Good to be here.