BBW25 - Day 1 - Agentic Economy: Autonomous Agents & the Future of Markets

All right, welcome everybody. Thank you, Ian. My name is Bob Bench. I'm the founder and CEO of Radius. I am very, very happy to be here with John McB uh from XDC and my very longtime teammate Anders Brownworth.

So, you know, I want this to be a flowing conversation. There's three of us. It's a it's a fairly uh open room here, so let's get weird. Um this will be fun. Uh what I really want to talk today about is um what problems is crypto solving for foundation models and the applications being built on top of foundation models and then alternatively what are foundation models and the applications on top of foundation models solving for crypto.

Um, aside from there being a lot of venture capital in both buckets and for founders like me really appreciating the fact that there's some VCs do want to sprinkle both in each pot. Um, so John, I'll start with you. You've been in industrial automation, you're now with XDC purely on the applying crypto to artificial intelligence systems application. Where do you see is like the sweet spot that you all are looking for? So I think one of the I guess low hanging fruit is going to be as we enter into a gentic economy where we actually have these agents pool of agents that are working on my behalf doing things that I would typically be doing dayto-day.

They need to be able to source information uh from the outside world. And so how do they get that information? This is all stuff that's kind of being built into existence in real time right now. But one of the primary resources that we have is MCP servers model context protocol and that allows agents to communicate with the outside world and leverage existing APIs for different services. And so if I have this pool of agents that are acting on my behalf, they need to be able to go and discover uh the services or the resources they need to do the tasks that I've given them.

And they need to be able to in some cases pay for those services. And I think that is a nice sweet sweet spot for how agents can leverage crypto as a form of uh seamless payment either agent to agent payment or agent to service payment via MCP discovery. Um that is currently being built up right now and there's room there for for crypto to come and play. I think there's a largest case for stable coins uh to come and play there. Then you see traditional finance coming as well uh with Visa and Mastercard building out agentic solutions uh that are going to use traditional rails.

And so this is all coming into existence right now which is extremely exciting. And one of my roles at with XDC is education and developer relations and how do we get people who are really well grounded in blockchain on the XTC network to kind of see this larger landscape that's coming into focus so that they can look uh for the solutions they want to build for the future. Um and so we try to expose them to you know this is the technology outside of your comfort zone. This is how approachable it is. This is how you introduce it into the applications you're already familiar with.

But I think that's one way that crypto can come and work with an agentic economy.

Specifically with XDC, how do you see XDC playing a role for the agentic economy? I honestly think that like I mentioned, stable coins are going to probably play the biggest role and we're going to see L1 tokens almost act more like gas than a native transfer of value. And we're probably going to see, you know, apps that want or agents that want to uh pay each other with stable take advantage of that 247 uptime that blockchain provides. uh be able to ideally in some cases use something like ARC which is circles new all one that they've created where you can actually pay for gas as USDC which I think is a really nice feature. Um, but one of the things XTC Network has just done is integrated with USDC.

And I think one of the main benefits there is going to be opening up the uh opportunity to build apps that hopefully have a little bit of stickiness to keep some of that uh USD flowing over XDC for agentic purposes. And that's going to be a struggle that all L1's have is how do they get a pipeline that keeps that business on their L1? you know, what is the the value proposition that's going to draw people to come and transact on their network. Um, XDC is focused on trade finance. They're focused on real world asset tokenization, but they still need to have these sticky DAPs that are going to keep the business flowing over their chain.

Honors, you've been looking at this now for, you know, we started radius looking at agenda commerce about a year ago after um, Romesh at MIT told us, you know, this is coming. What are you building? Um what are the unique like say say you have a stable coin dealing with agents? Um what are the product requirements that are unique for agents as opposed to say you and I flinging some tether at each other?

Yeah, that's a great question. So um it's counterintuitive. Uh you might think that okay I go to an AI agent and I'm going to say hey you know book me a trip to Singapore whatever. Maybe it buys you your plane tickets, your hotel. Not really.

Like it does, it can, but that's not really aic payments. What it really is, in my opinion, is uh you know, okay, so we're going to make a trip, so we got to find out, well, what are the places you want to see? What is the weather likely to be, etc. All of those things, uh, you know, right now, some of those things are of lower quality available on the open internet for free. But as AI agents really scale, they're going to end up, you know, pretty much making that ad business not tenable anymore.

And those data sources will, you know, cease being so free and probably become a little bit better. Uh so if I can buy an agent can buy uh the weather forecast or can find out what the weather is right now at a certain place etc. that's the kind of uh you know agent commerce that that uh that I see coming. It's really not a you know Let's do this large purchase, purchase this ticket, purchase this hotel. It's really all of the little bits of information that I need to look up.

Well, when is that museum open that I wanted to see, etc., etc. All of those tiny little things. Your one AI request might end up in 10, 20, 50 small kind of microtransa transactions that coales the data necessary to make the overall, you know, better output the decision. So, it looks very different.

It kind of, you know, It doesn't look like something where I take a USDC and simply say that you know I'm going to use that to pay for my plane ticket. It's really something that uh you know really looks a lot more like uh microtransactions I would say.

So so kind of to drill down on that. So I go to I go on claude and I pull up my MCP server and I say all right I'm flying to like fix my trip to Singapore. to your point, that's not one transaction. That could be 50 or it could be a thousand depending on how many requests that calls for from data.

Yeah.

How does that happen when your ETH transaction costs you four bucks?

It doesn't. That's a good uh it's a good little segue there. Uh so, right, it really does need to operate on the uh you know, at least under a penny uh per transaction or else this whole thing is kind of you know, dead on arrival, right? So uh so you know the ideal system here would be something that is able to do you know extreme volume and you know we were working at at the Federal Reserve on a on a central bank digital currency. Turns out that for the dollar there's you know 8 billion users of the dollar.

It's like there's a lot of people and those eight billion people could make you know certainly hundreds of thousands probably millions of transactions per second at peak. So you need a system that's able to be in that range and if you have some kind of limit where you say well look we can get up to a million or and no more that's that's really not going to win right so you need something that can scale but before you need to scale you also need to be able to do that at you know tenth of a penny hundred thousandth of a penny kind of levels or else this whole thing really doesn't work. Uh so you know that that's the technology that we've been working on for the past few years now and It turns out that uh counter to popular belief, you can actually do that can have that world. You can uh you know make a system that that runs in the multiple millions of transactions a second and and clear payments for you know way way less than a penny.

So kind of to summarize one way to handle the egetic economy question is you go really fast and really cheap in a step change. Um that's slightly different than trade finance and RWA you know as far as the XDC build what are the unique requirements you know in addition to circle saw you guys kind of one of the first chains to adhere to ISO 222 like what do you think when you talk to users the black rocks of the world the Maris of the world who do those types of of of products what are they asking for from XDC that's going to help that future of automation?

Good question. Um, I'm not typically the person talking to them, but uh, I think a lot of how you get into those conversations is who you have in your network. And traditionally, that's been one of the stronger points for XTC and those and especially in trade finance is it has people that are in traditional trade finance that are already participants of the XGC network and that's opened doors, has made connections and so you have that trust um, for adoption, you need to be able to solve a use case. And so we see um things like there's a TRA token which was invoice factoring on chain. Um so there are use cases being built up to be proof of concepts to show that these things can work on blockchains that they're uh trusted that they can be adopted and I think we need to keep building those out.

Um I don't know what triggers adoption like on a mass scale for XTC in these world in this world but it's something that we're trying to unlock. So with back to you Anders on as we think about agenda commerce as it kind of is growing at least in conversation at least on the radius side we're seeing there's a lot of white space needs to be built there's a lot of conversations not a lot of builds yet that are ready for ready for the real time uh on the risk side like what do we need to be worried about like we you know we launched say you know they launched on USDC say we launched on something like USC What do we need to worry about the risk side when agents start moving our moving our coins that we weren't ready for when we launched USDC back in 2017?

I don't know how to answer that. There there's uh you know it's a completely different world from when we launched USC to begin with. You know, we launched USC on on the Ethereum chain only at the time. There's a few of us. It wasn't like a you know large operation like it is now.

Um so uh you know some of this kind of hearkens back to the the last uh last interview that was going on previous uh is you know the those uh you have to like look at the risk framework of all of the systems that you're using to clear the money right back in the day when it was just Ethereum that was just that was a very specific thing is I would argue probably pretty quantifiable it's a lot harder now um so uh you know you could talk about what is the singleness of of money look like across all of these systems. Um, you know, you have some systems that are public permissionless, incredibly neutral, you know, low transactions per second, high cost, but others that are, you know, you trade some of that for a bit of centralization, a little bit of, you know, a little bit less on the on the fees, a lot faster, things like that. So, you you have this giant spectrum. It's a lot harder to model and figure out like, you know, what is what. So then you know the the open question I guess is is a dollar on this system exactly the same or transferable one to one to another right the the primary dealers the issuers of stable coins would say yes they are because they're able to mint and burn on any chain so it doesn't really matter but but that's not how people evaluate stable coins they're evaluated on the the uh secondary market uh you know what what can I trade some tether for some USDC for on these different chains etc.

And that's where you'll see the, you know, the the risk expressed in the difference in price. I guess

what what type of risks do you all see on the industrial automation side for application of stable coins?

I mean, typically with industrial automation, we're looking at not so much moving value over blockchain as uh information, right? So, and it doesn't it's not as applicable to this conversation, but I'm looking at ways to uh get inference closer to the data source so that we can try to predict say when a set of pumps on a system is going to fail based on how they've acted in the past. Um that's kind of what I look at and why I look to bring uh AI into that world. Um doesn't directly translate to the blockchain conversation though.

All right. So I'll I'll ask a basic question you Anders is why do we need a blockchain for this?

Well, you you uh you want to go to a blockchain when you want credible neutrality. I think blockchains sell neutrality, right? They sell uh you know essentially a trust layer. Uh you would want to not go to a blockchain when you don't want that uh uh you know that you want to pay for that, right? So when would you not want that?

Well, you know, if you're buying a house, you want a blockchain because you want the irreversibility, you want the uh you know that that that trust that you get. Uh but if you are buying uh a weather forecast for a penny, that's okay. You know, like I actually I don't need a blockchain because I don't want to pay for the reversibility of a penny. Like that that doesn't make sense. Like I I don't want to have to pay multiple pennies to move a penny.

that's now a not profitable operation. Uh so you know basically what I'm talking about is is as volumes go way up and and the value per transaction goes way down you don't want a blockchain. Um as the values are high and the the throughput is is not like you know into the millions of transactions a second you want a blockchain. So

thoughts on that John? Blockchains versus non-blockchains for moving value on artificial intelligence systems.

I mean that's uh that's really well said. Um it makes sense to me. Uh so and actually I never thought of it the way you explained it first off where a lot of agentic transactions will redefine how we actually think of transactions in general. Uh because you're right, demonetizing the web by not caring about advertisements, moving most of the internet transaction to be agent to agent instead of human to website uh changes how the web has to work. And I'd never really considered that you need these microtransactions.

You need a market around how you can even go in and get the data that you want and you need discoverability of that data. So you need almost an entirely new technology stack uh to enable this agentic economy. It's really interesting.

So I think kind of going forward that thesis of value on artificial intelligence systems and we've talked about discovery of data, we've talked about discovery of inference, markets for those things. You know, one thing that happens every time there's a major technological change is incumbents get displaced often very fast and very violently. I guess I'll start with you, John. Like, who is the company with 50 60 70% operating margin in the S&P 500 that becomes a lot less important in 10 years from now because of this technology?

I I think you could pick a lot of these. Um, I think that you're going we've already So, creating a startup, I'll start with a small scale. Creating a startup today is very difficult because most ideas that I've had for how to apply AI to what I want to do. Uh they end up getting written by the Frontier Labs and rolled out to me virtually for free. The cost for the subscription and now you're seeing things like the way that the internet's monetized.

That's changing. OpenAI is talking about building their own browser. Uh you see things uh traditional uh uh business tools like Word or Excel. Uh those are going to be changing too. And it looks like we can start to build that uh into an open AI offering as well.

And so there's huge disruption. Um last month my 8-year-old daughter vibe coded a game uh for for the web. And that's crazy to me that we're at the point now where you can have bespoke software. You have an idea for anything, a game or a product, and you can create your own on demand when you need it, and you can use it. And that disrupts a lot of how today's economy functions.

When I have a desire, and I can just bring it into existence with the help of artificial intelligence, um, it makes it a lot more difficult to have those billion-dollar companies unless they're building out the infrastructure for this AI.

Yeah. You know, I suppose there's two ways that this could go. One is we could, you know, watch the whole thing blow up and kind of forcibly rip over to a new thing. Or we could have an orderly transition where one system becomes less and less used and important and it sort of fades off into the sunset while another kind of picks up the slack and goes. I'm much more interested in that.

I I'm not a burn it all down kind of guy, right? So the the question I don't think is like who gets drastically upset? Some probably will and I'm not good at like forecasting the future. I'm an engineer so I don't do that well. Um however uh the the I think the long term is you know the the question is really how do you infiltrate and how do you turn the ship on some of these large operations that you know essentially will be disrupted if they don't do something.

Uh that that's that's the more interesting play in my opinion.

Yeah. I think um the most noticeable one I've seen in recent history was um not that people like Anders and I are watching Circle share price very closely as foreign employees but at one point Circle uh was uh 60 no was 30 times the valuation of Western Union uh Western Union still was about 2.5 billion in market share and uh Circle at one point what I I want to say 60 um and Western Union has physical offices in every country in the world and 17,000 employees Uh so you know I think a lot is coming definitely for the people who move money uh and the people who move information. Um before I go to questions because I like questions they're fun. Um this is partially an AI conference.

Uh honors I know you have strong feelings so I'll let you start on this one. You know that I'm team anthropic. I reach for claude every time. What what is your default when you reach for a foundational model system? Which one you go to first?

Open models. I uh I I don't like the uh privacy implications of a chat, GPT, Anthropic, etc. So, I I use open models for most things. Just run run it myself. Um that doesn't solve everything.

Oh, those are some great coding models, by the way. Uh doesn't solve everything. Works great if you're on an airplane. You don't have an internet connection. They haven't gotten Starlink yet.

uh you know that that's fantastic but uh you know so you have like a couple of features but really the problem is uh is privacy I I don't want to leak information

John

I agree uh so I I tend for accessibility um when it comes to education uh something accessible uh so Gemini I use nano banana for image generation and Veo3 have been mind-blowing to me just the multimodality that's coming out right now is insane. Uh where I can literally generate any video or image that I want exactly when I want it. And so I spend a lot of my time with Gemini. Um for work I tend to do more with OpenAI and with their APIs. Um really that's more of a business decision though.

What what did your uh daughter do the her game in?

That was OpenAI. So that was when they introduced GPT5. um just coding through their canvas to make a single file application is so simple and then iterating over it. Its ability to remember what you like, what you don't like, and to only change the things that you want changed was was new to me. Um typically you have all sorts of trouble when you start to iterate with these things.

And so I was really impressed with GBT5.

Interesting.

All right, so we have about five minutes left. We'd love to hear questions if you all have them. Um, the three of us have been in this space for a while. Uh, I think I read the Bitcoin white paper about a mile from here in 2010. And I tried to start a Bitcoin practice.

There wasn't a lot of Bitcoin legal work in 2010, but I I tried that back in 2010. So, any any questions?

Will SAS be will SAS be obsolete in the future if we can program immediately?

All right. Will software as a service be obsolete? honors.

Uh no uh no I I think uh what will be obsolete though is a single so large software code base used for everyone that'll be that seems that's on the chopping block. If you have a system that can effectively dynamically change per customer then what does that look like? giant system where every individual user is treated completely differently as per their requirements or requests etc. So can you sell that as a service? Yes, absolutely.

So I no I don't think so at all. I

I think one thing that I've been most questioning is the sales model of software as a service. The fact that these organizations spend a incredible amount of their capex on sales revenue or actually salesperson's compensation. I think that is going to be dramatically changed. Um, other questions. So, so, uh, Cubic Coin, uh, recently did a demonstration, they called it, um, when in taking over, you know, they claim they took over, um, close to 51% of the Monero uh, network.

Um, what are your thoughts on that? Do you think that, you know, we should be concerned about the capabilities of AI and blockchain uh working together to um attack other networks um blockchain or blockchain or non-blockchain? And do you think you know how do we protect um or do we protect you know uh historical coins right like Bitcoin and Monero from uh those type of attacks?

Yes, we love the conversation 51% tax. Anders and I have spent a lot of our lives thinking about fifth warm sand. Andre, I'll let you hit this because you've lived it much more closely than I.

Yeah. So, uh, you have, uh, should we be concerned? Yes, absolutely. We need to understand the risk framework that you're working under with a blockchain, the, you know, consensus being overrun with a 50, you know, 50 plus percent attack. That is a that is a thing, right?

But what can you do if you do that? Can you, you know, uh, create a new transaction that spends somebody else's money? No. Can you include that in the canonical chain? No.

You can, but you know, it's not going to last. You have to keep you you have to keep the attack in force. So, no. You know, if you could get a complete capture, then yes. But, you know, it's it's a constrained set of things.

You've seen a number of coins that kind of become 51% attacked and then and then they're not and then they're 51% attacked and then they're not. That's not a great place to do business. We have to be very concerned about that for the sort of the majors I'll call but uh that's that's not sort of the end all be all, right? So the the other consensus that you have is effectively not a distributed consensus is a centralized system, right? And that has a whole series of attacks too.

you can, you know, come in and and like take out whatever that single player is with a with a, you know, a hack or something or you could go after them from a a legal perspective or something like that, right? Because that's the that's the ecosystem within which they're playing. I think the important thing is go in with your eyes open. Understand what the consensus mechanism is of everything that you're using. Even if it even if that consensus mechanism is lawyers and the court and uh you know sort of judicial process that is a consensus mechanism for the permission space.

So you know you just have to like look at what they are and understand what they are and you know should we be concerned when they get yes we should absolutely be concerned because these these things provide capabilities that you can't just you know get anywhere else. So you don't want the very top the majors to really suffer from something like that. But yeah, I mean that it is possible is true on the on the you know the the permissionless decentralized side and also true on the centralized side just in a different way. One thing I love about 51% attacks, and I I do think they're a feature, not a bug is um you learn real fast who's actually decentralized when a system gets attacked, right? When you can reorder your chain in a matter of hours, I strongly question whether you're decentralized.

If BTC gets 51% attacked, which it really just it's attempted attacked every day. It's just it's a really large mining pool and it's really hard to do. Um, if if if someone succeeded in 51% of attacking Bitcoin, it'd be really hard to undo that because it is generally that decentralized. I mean, Coinbase would have something to say about it because Coinbase is holding a lot of the world's BTC. Um, but that's interesting.

Um, but I think we rolled back in 45 minutes. Um, Salana's validator test was mostly through one validator in Switzerland, right? Um, I think the cool thing said is that if it actually works, it proves the thing was decentralized and that kind of approved Monero was mostly decentralized. Vercoin was decentralized because that went through. Um, but if code is law in the sense of blockchain, in the blockchain world, then you need to uphold the law and don't reverse it.

That's my view on it. Um, so it's it's I love I love the fact that 51% attacks exist because it actually shows whether you're decentralized or not before standard

in in DeFi. Circle is the centralizing force in DeFi.

Correct?

Because if Circle says I'm not going to redeem on this chain instead I will redeem on this fork of the chain then that fork of the chain becomes the true chain. So how D is your DeFi

and and we're at time but I think that's good any point which is uh you know artificial intelligence systems whether it's foundational models and applications on them are going to give you offensive and defensive capabilities there usually the um the attackers start start the fight and they're pretty good and I think we have a lot we'll see a lot coming up but I think it's exciting exciting for the future. I want to thank John I want to thank honors. Thank you all for being here. Thank you Ian Kane. Thank you city of Quinsey.

This is great.

Thank you Bob. Thank you.