Intent-based transaction systems like Near Intense enable seamless cross-chain asset swaps by abstracting away chain-specific complexities, allowing AI agents to execute transactions across multiple blockchains without understanding underlying technical details. Privacy is becoming a critical primitive for the agentic economy, with implementations at asset layer (Zcash's Orchard), smart contract layer, and chain level (Starknet's privacy pools) working together to create end-to-end confidential ecosystems. The convergence of intent-based execution, privacy-preserving AI, and cross-chain interoperability represents a paradigm shift toward more accessible, secure, and private digital financial systems.
Install our extension to search inside any video instantly.
Virtual NEAR Day
Added:[music] [music] Heat. Heat.
[music] >> [music] >> 17 floor.
[music] [music] We'll see you around the timeline.
Feels good to be back.
I think the [music] rollup is one of the most premier media brands in crypto and Andy and Robbie are [music] great experts at everything they talk about and it's my favorite podcast.
>> Adding it to my library every >> potentially multi-chain deployments. We are at the connect [music] day one here.
I'm guessing from base.
[music] [music] [music] >> [music] [music] [music] >> Heat. Heat. N.
[music] >> [music] >> JP Morgan, Black Rockck, DTCC, Fidelity, the entire thing was just institutions.
It's just it's next level. This industry is going to the next level. And guys, I I don't know what else to say.
>> I'm bullish.
>> He said it. He finally said it.
>> I'm so bullish. And he's bullish. I'm bullish. We're bullish.
[music] [music] Heat.
[music] Heat.
[music] Heat. Heat.
[music] >> [music] >> Corola is headed [music] to the top. If we're not already at the top, we're going to the top.
>> Roll up is my favorite.
>> People from all walks of life, all over the world, all over this industry come together for the roll up.
[music] [music] [music] >> [music] [music] >> Zoom out. The institutions aren't coming. They're simply here. It's the golden age of crypto, guys. Get right.
Let's go.
[music] >> Welcome to the next. See you on the show.
And we are live. Welcome to virtual near day, guys. Near is our exclusive AI partner here at the rollup. It is July 16th and we've got a fantastic show for you guys today. Uh we're doing a bit of a stream to takeover today. You can see up in the top corner, this side over here. Powered by near. That's what we have going on. powered by Near Today, our exclusive AR partner. Obviously leaning on the intense front and the AI front, guys. Uh I'm excited. Been uh planning this for a long time, and it's finally here. And so, we've got a fantastic speaker lineup for you guys today. You guys heard from Alex Chevhenko yesterday. Uh and he is the co-founder of Near Intense. Near is bouncing off the intraday lows. Uh we had a bit of a neutral morning and now we are looking at what is going to happen the rest of the day. Appreciate you guys rocking with us. We're going to get into everything that we have going on. Hit us up in the chat. We've got a tremendous lineup. As I said, we're going to start off with Matt Henderson.
He's going to be talking about the app that ate the stack these apps are eating. Uh and we're going to follow that up with Harshit of Near Intense. uh Hugo, Kyber Swap, Felix. We we got a whole uh virtual conference for you guys here today. So, it's going to be an exciting time. Uh we've got keynotes, uh we've got panels, we're going to keep it spicy, we're going to get debates going, we're going to have a great time. Uh we're going to be chatting about apps, we're going to be chatting about intents, we're gonna be chatting about near AI, we're going to be chatting about confidentiality. You guys heard about Alex and his confidential intent story yesterday. I mean, also, you guys probably heard, uh, there was a guy in the Near community that made a made a rap song for us. So, we're having a lot of fun. Uh, we have been big proponents of Near. Remember, Near is one-third of the three-headed dragon. There's sort of this uh this I hesitate to call it a narrative. feels more like a it feels more like this uh paradigm shift towards fundamental and and a return to the cipher punk ethos of the space. And so we've obviously seen what happened, you know, to Bitcoin with its ETFs with Micro Strategy. It's become a bit institutionalized. All of this has been taking place. Um, but there are projects, companies, coins, networks that have actually embraced what the spirit of the space is, what it's always been. And I like Near because I consider it one of those companies. Um, it's very, very cool to see what they've done. Uh, Near Intense has done over $23 billion. There's a couple of dashboards that I like to point to. Uh, I also before we get started here, I want to pull up and I don't I don't think we're going to get it up on the screen, but I saw a dashboard not long ago around uh near uh the buybacks that are happening.
They posted um kind of this near tokconomics sort of dashboard uh which I found very compelling. Uh I think it was very very cool to see. I think more uh projects and companies are headed in this direction. We also heard from Alex Chevhenko that they are uh doing what they can to reduce the inflation and so I think all of these things you know if you look at the real company uh yes so this is what I was looking for look at the key metrics the expanded nearrevenue dashboard is now live uh and I want to just call out a couple of these things from the nearrevenue dashboard this is at revenue.near.org or uh they are closing in on $40 million uh of revenue.
Uh this is updating in real time. Uh and so intent volume is upwards of 23 billion. Uh confidential TVL is at 33 million unique users over the last 30 days, 550,000.
Monthly net revenue is closing in let's see monthly is [snorts] >> monthly revenue [clears throat] >> 253,000 revenue as percent of near emissions is around 10%. So very very cool to see all of the revenue that's taking place um you know as crypto becomes a little bit more legitimized you know we're seeing it enter into uh the more robust ass it's really treated as as a real asset class right and before I think we could kind of get away with these you know network state Jevans paradox compounding effects and uh this was going to take us to Valhalla but it turns out these are companies just like all the other companies that are out there, they should be valued as such. And so that means we're going to look at their revenue. We're going to look at how durable their sales are, how durable is that revenue, you know, what is the market cap to revenue ratios. We're going to be analyzing all this because at the end of the day, these are real companies and they're going to be valued as such. Uh shout out to the guys in the chat rocking with us. Synchro Path, uh Ashro, future is near. It certainly is.
So we called out some of the intense volume metrics. We called out some of the revenue metrics. Uh I think it's about time that we get into our content for the day. Our first guest is going to join us in the back room any minute. Uh our first guest today is Matt Henderson.
We're going to do our best to stay on track here. I think we've got about 15 minutes or so. Uh about 20 minutes. So uh and then we're going to get into our panels. But first, we've got fireside with Matt Henderson. He is a builder uh in the near.com ecosystem. Uh near.com obviously is this uh combination of several facets. They've gone and I think taken a path that a lot of uh projects are now coming around to where you know you've sort of got this disperate ecosystem of apps and projects.
Everything's very decentralized and this decentralization kumbaya. You know what I think near has done is they've taken a very practical approach to verticalization. They've said, "Look, we're going to bring these things under one roof. It makes sense to have an intense protocol right next to an AI protocol. We're going to see why over the coming years, why this makes more and more sense. I anticipate more projects are going to come to this realization. Uh, and they're going to start to unify a lot of these user experiences to ultimately provide the best user experience for the consumer.
So you guys heard you know 23 billion plus in intense volume uh on near intense uh that is getting over to near.com. Uh and then you've got all the great things around AI agentic economy agentic commerce that are happening on near.com as well. Um truly competing on both the financial side as well as the frontier technological side with respect to AI. Our first guest Matt Henderson, I can see him in the back. We're going to bring him up on the show. I think it's his first time on the show. Matt, it's a pleasure to welcome you. You are our first guest for Virtual Near Day. How you feeling?
>> Good. Glad to be here. Pretty happy to be here.
>> Absolutely, man. Well, it's a pleasure to have you. Um I think it is your first time on the show. We're going to be talking about the app that the stack meetne.com. Uh you're a builder in the Near ecosystem. Because it is your first time on the show, I'd love to, you know, just understand a little bit more about you, you know, and how you ended up uh getting involved with the Near ecosystem.
Um, okay. Yeah, that's a that's a that's a story that goes uh way back. So, um yeah, many years ago, back around 2010 or something, I was running um a product agency and um I hired some a couple of very young Finnish brothers and one of them was a deep cipher punk and he was um he was one of the first 1000 Bitcoin miners and uh yeah, he was so he was trying to encourage me to mine Bitcoin, but he was also uh developing his own uh floating country protected by autonomous weaponized drones. So I didn't tend to listen to much to to uh to what he was into. Um but yeah, a few years later he kept on and kept on and we had some people some remote people working for us and so um they wanted to be paid in Bitcoin. So that kind of got me more interested in Bitcoin. And then um in the DeFi summer of 2020, that's where you know everything just kind of uh kind of you know I just went into the rabbit hole seeing uh Ethereum and smart contract platforms coming out and seeing you know that that an entire permissionless financial world can be built on them. And um that same Bitcoin miner, he went to work for Near and he invited me to uh to join him and we were working on a on a project called Aurora, which um is a it's like a it's like a blockchain implemented an EVM blockchain implemented as a smart contract on NE.
So that was really cool. Um and that's how I got into the Near ecosystem. Um once that project was up and running, I um I I left kind of my do my to do my own things. But when I saw Near Intense uh coming out, I uh I contacted Alex Shvchenko, our old CEO from Aurora, and we were talking that there's really a possibility that the technology is right now to build a retail app. Um that's kind of category defining. And so that's how I ended up back here.
>> Yeah. And we talked a little bit about the the stats that Near Intense has been able to rack up up to this point. $23 billion in volume. I think upwards of 40 million or so in in you know fees and and monetization and revenue uh 500,000 users you know the numbers go on uh and we've seen the chain uh you know the crosschain ecosystem the bridge ecosystem evolve for a long time but this feels like you know you guys have been able to crack this thing and unlock ecosystems uh that you know previously were walled gardens zcash is uh you know obviously the name uh that comes to mind there we We had Alex Shvchenko on the show yesterday and so we got really really deep in this. Uh Matt, what I'd like to chat with you about today is how you're bringing this into the fold because Near.com is sort of this you know overarching umbrella container for several different things. Near intense obviously plays a very significant role in that. But maybe you could talk to us about how near intents interacts with some of the other components around near.com and and ultimately what is the vision for this you know all-encompassing platform that involves you know AI near intense g could you just give us sort of the overarching view of near.com and how nearense fits in?
Yeah. So, um the the vision for near.com is well, we we kind of feel like the term crypto uh or we're kind of in a transition phase where crypto as a term might be fading a bit into the background. Um just kind of the same way that no one says, you know, we're an internet company anymore. Like all companies are internet companies. So, it feels like that kind of the blockchain world and the financial world are are merging and and very well integrated to enable this kind of application to come out. So um so yeah in in.com um our our our vision is one financial app that you want to that you want to live in and so um we kind of think of eating the stack in this way uh is is kind of uh deleting various stacks of friction in your life.
So whether that's swapping, so in near.com since we're built on near intents, you just see all of your assets in one place, whether it's Bitcoin, whether it's ZC, whether it's uh USDC, you don't really care what what chain they came from. It's just all there together. Um and in the future, we'll be having real world assets, ETFs, stocks, uh bank accounts, and uh it will begin to just Yeah, you won't really know if you're dealing with a crypto app or or a real world uh you know, financial app.
Yeah, it it is pretty incredible to see um the rise of tokenized stocks, stock tokens. You know, we've seen Robin Hood chain. We've seen how, you know, Near Intense can help unlock these ecosystems. Um I was chatting with someone that was representing Robin Hood chain and they talked about this mechanism where, you know, obviously there's a lot of equities on Robin Hood and they're able to just sort of drip out these tokenized stock versions of that when someone submits an order. And I could really see a vision where you know near intense helps bring those those equities that may result or excuse me they may originate on something like a Robin Hood chain because of the proximity to the broker but near intense is sort of the you know the pathway or the mechanism that the vehicle that brings these things and proliferates them all across the onchain ecosystem which is a very very powerful tool because it's able to take these assets exactly where they want to go. But as you're talking about, all of this can live in the near.com interface. And so maybe you could just, you know, obviously I think the the thing that people are most familiar with is going to be buying, selling, swapping, you know, via near intents. I've started to play around a little bit with the near AI. I think even before it got uh subsumed into the near.com interface.
could you just you know maybe start to tell us about what some of the features some of the capabilities are on near.com as it pertains to AI and how you foresee the agentic economy progressing in this regard.
[clears throat] >> Yeah. Um well right now we don't have AI features in the application because we are doing a lot of thinking about how to um how to best integrate them and um and I think it's it's very difficult because this is obviously a technology that's going to change everything but it's it's very difficult to look into the future and say you know what is that going to look like and so we're [clears throat] trying to take some of our own experiences uh for example everyone in the team are using AI agents on their own now and um I can give you an example. Uh a couple of weeks ago I was out on a bike ride and I realized I needed a charger for my iPhone and so I just pulled off the road and I had built I had built a skill for my agent uh to do shopping online at at Amazon. And so I told it I just told it I said, "Hey, I I need a charger for my bike. Um I need you know it's got to be thin because it's got to fit into this bag."
I sent it a little photo of the bag it's got to fit in. I said, "I don't want a cheap off brandand. I don't want it to explode while I'm writing." And um it went off. It found some products. Uh it couldn't determine which was the best one just from Amazon. And so it went off and researched it and it finally came back and said, "I recommend this one."
And I said, "Okay, order it." And it said, "All right, it's ordered. Your email is on the way." And so that was really a magical moment. I I didn't think I'm using AI. I did I just thought I'm I'm expressing what I want to do and this thing has you know it it I don't even know what systems it's interacting with. So we're really thinking about uh AI in terms of how would you want to interact with a financial account um when there is intelligence there. Uh so that's you know without giving too much away about what we're thinking that's that's the direction we're headed. So it it would be much more than just like a chatbot that we would add to the application. We really want it to be able to do things on your behalf. Uh we want to integrate long-term investing into the application and there's a lot of education around that. There's a lot of automation around that uh that that having intelligence can do for us.
>> Yeah. and and you know this is sort of this pendulum approach where you know the there are so many elements and education is part of this and and you mentioned so many different use cases here but simplicity is one of the things that we've continued to lack when it comes to you know these onchain applications and and you know the thing that you guys have been able to do and I think part of the reason for the success of near intense so far has been because of its simplicity so you know given that you want to tie in several of these different features you know it to to build something not only that goes across chains as Near Intense does, but also, you know, is capable of um tailoring itself to different time horizons as you mentioned. You know, people might want to buy and hold versus others that might want to trade more frequently. H how do you embed simplicity and keep that in mind as you're building out such a varied feature set?
Um, I think it's uh it's very I think it's very difficult for builders in our space to get to that level of of um of simplicity because it's very hard for us to uh deeply have empathy for the audience that we're building for which is not necessarily the audience that is here today. I think us in the crypto space as a as an audience, you know, in terms of the use of blockchain technology is going to be shrinking over time relative to the rest of the world that's going to be getting here. So, having empathy and be able to put ourselves in the shoes of others. Um, also recently I've onboarded uh I tried to onboard a friend into crypto. Uh, very smart person, a doctor. Um, and I really this the first time I've had such an intimate view into what it's like for someone who's never experienced the space but is interested in it to get in.
And you're dealing with wallets, you know, web three wallets. Why does Rabbi support 140 chains and then but for Salana I've got to have this other chain and for you know this other wallet and you know etc etc >> and um I just you know with that experience I thought oh my god you know we take for granted so many things that other people would have no idea about and so um that's just been a tremendous uh focus in in the building of near.com whether it's um kind of blank slate screens when you go into an area that's not been used yet to explain what that is. Um, not using jargon. I mean, we we put a huge effort into not having jargon anywhere. Uh, what is staking? You know, how do you communicate what staking is to someone who has nothing to do with crypto >> um or you know, lending all that kind of stuff. So we put a huge amount of effort into the language around these things and trying to trying to bridge a new paradigm with the existing context that people have been used to living in for all you know their their whole life.
>> Yeah. And you know our our world the language that we use it feels very distant and foreign to a lot of people.
And so I think finally people are realizing like okay it's best for us to meet them where they are rather than expecting some of these newcomers to come in and you know educate themselves and and you know kind of overcome that high barrier when it comes to the language and the culture and all this.
let's lower the barrier of entry, you know, and and meet them where they are so that they can also take part in these tools, empower themselves and and and I think you're, you know, you're absolutely right to keep that in mind as you're using that those principles in everything that you're building there at near.com. So, so as we look forward, you know, when we think about what's next, you know, we mentioned sort of this next cohort of users. Obviously, there's sort of a closed, you know, there's there's a group here that have been here for a while, but the idea is that we're going to expand the user base significantly in terms of humans also in terms of agents that, you know, could be using this technology and these tools. How much are you looking at, you know, building with agents in mind? Do you foresee maybe you know one interface just interacting with agents and then giving you know the agents the ability to transact and giving them all these tools as you think about what's next you know forne.com real world assets fiat connections everyday financial tools what are some of these things that you're most focused on keeping in mind when we look at you know the next iteration of near.com >> um well in the in the long term as pertains uh agent authentic activities.
Again, I just kind of look at at at what my own experience has been and my own experience is um integrating more and more agents doing things on my behalf in the for me and seeing more and more of the products that I use in the world. I'm getting, you know, they're they're letting me know, hey, we have an API now or we have an NC MCP now that your agent can work with. And um I was kind of thinking what is what is kind of like a crypto you know analogy to this and we've talked a lot about chain abstraction and now I'm starting to think about this term product abstraction. I just want to tell my agent what I want to do, what I want to get done and I don't think much anymore about which of my products the agent is interacting with. So we would definitely want to have a means for agents to be able to interact with your near.com account. And then we would want to make near.com the right place for all the all the financial things that you would need uh to live. We would like that to be you know no better place than than near.com.
>> Yeah. And I I think this is uh you know it's becoming uh the case. And so as as you're developing you know we also understand some of these network effects right like liquidity begets liquidity intelligence begets intelligence. you know there's data that is going to be involved in both you know at the transactional level as well as at the prompting level all of this can facilitate you know a better building experience and feed into one another so that all these products can get better they can all help them you know one product can help another and vice versa um Matt fantastic really appreciate you helping us kick off the show just want to kick it over to you give us any any insight you know any advice that you would share with builders that's a closing remark any closing thoughts that you have that we can keep in mind for the rest of the uh the rest of the show today.
>> Uh advice for builders. Um I guess the advice would be to think in terms try to think in terms of outcome.
Um getting paid, sending, saving, ordering, um that kind of thing. And try to as best you can decouple yourself from the technology that you're that you've been used to all this time. and um and really try to have empathy. I think that is the the the hardest part.
It's so easy for us to just say like the word slippage and uh and you know that just rolls off our tongue. We all know what that is. But if you you know if you talk to someone outside and you show them slippage, you might scare them.
They might not finish their their [clears throat] transaction. So just think about those those little details.
>> Yeah. Fantastic. Matt, thank you so much for joining. uh rockstar first appearance on the show. Looking forward to uh having you on again soon and uh continuing the rest of our virtual near day.
>> Thank you. Thanks for having me.
>> See you next time. Thanks, Matt.
All right, guys. We're going to keep things moving along. Fantastic first guest. Great way to kick off. Uh you know, you heard a little bit about the app that ate the stack. Near intense, you know, was such a significant portion driven so much volume, but this is just one component to the overall stack here.
uh which I think speaks to the volume at which Near is truly doing. Uh every crypto app promised to simplify onchain.
Near.com is actually doing it. Swaps, pers, payments, every asset, one interface. Do we get into a place where you know the interface is just you know the interface for the agent? I think it remains to be seen. we could see this chain abstraction, chain aggregation, product abstraction, product aggregation interface in which, you know, the LLM, the agent is the one that then goes and clicks the buttons, pulls the levers and executes the transactions. Uh, but it was great to discuss with Matt uh the future of Near.com. You guys have also, I'm sure, seen our episodes with Ilia where he talks about, you know, what it is that they're building for. As you guys know, if you've seen those episodes with Ilia, you know that Near uh is an AI first company. All of the things that they've done in the blockchain world, the intense crosschain world, all of that is sort of a side quest. Like, you know, Ilia started, he said he wanted to build, you know, AI. Um but he quickly realized that AI you know has a lot of valuable components and there needs to be this uh mechanism of value transfer so that AI agents can interact and transact with one another. Uh and that is what ultimately led to the near blockchain. uh there was you know quite the uh quite the technical depth when it came comes to sharding some different blockchain architecture design choices that were made and so that is what we're we're seeing live in practice today and all of these you know techn technological innovations can be applied uh to all of what we're seeing out there today uh guys I appreciate you rocking with us in the chat virtual nearday on the rollup circle brave starknet and more. Super excited for the insights and the announcements. We appreciate you out there. Hello Near. Give me some GM near in the chat. Stacked lineup. Circle swap. Excited for the announcements.
Here we go. Hello. Hello. Uh shout out Kings Clips. Uh shout out Ro Heat.
Appreciate you guys rocking with us.
Flippa 001 Robin Hood Near intense sounds great. Yeah, I really do think, you know, we've got this uh stock token uh tokenization ondemand sort of mechanism that is just dripping equities into the onchain ecosystem. You guys have seen, you know, BNB and Salana have a lot of stock tokens just because they've tokenized like this big inventory, but it requires them to just have this this bulk of inventory that they tokenize on their own. You know, I think with the tokenization on demand from Robin Hood, we're going to see these equities just drip out and then near intense is the mechanism or the vehicle that can then proliferate those tokenized equities across several chains. Uh let's see, Rag in the chat.
Appreciate you rocking with us. Near's work on resharding confidential execution rising intense volume is actually building real info for private agents. Yeah, it sounds like you've been watching our show. Uh, shout out to King's Clips Fire 142 in the chat. Wow, you guys are rocking with Near in the chat. Hit us up with questions that you want to hear our panelists answer. Uh, we're going to bring up our next group uh in just a moment. Tit title of this panel is the change shouldn't matter.
Remember this is virtual near day. This is essentially a virtual conference. I see our next group of panelists uh in the room. We're going to bring them up in just a moment. what we're going to be talking on this panel. An AI agent doesn't know what chain it's on. It's a pretty heady thing to think about. An AI agent doesn't know what chain it's on, and it shouldn't have to. Probably doesn't care what chain it's on. It cares about outcomes because ultimately that's what the human wants. It cares about outcomes. It wants cheaper, better, faster. It doesn't get involved the tribalistic nature of a lot of the culture wars that we've seen on these chains. The promise of intentbased execution is that agents express outcomes and the network handles the rest. This panel gets into who's actually building that infrastructure and what it takes to make those promises real. The chain shouldn't matter. We don't have Harshet from near. Uh we've got uh we've got our friend Tyler filling in uh with him. Uh Tyler Bond is gonna be replacing Harshet. And we've got Felix from Cow Swap. We've got Hugo from Kyers Swap. Let's go ahead bring him up on the show and get into our next panel. The chain shouldn't matter. Guys, welcome to the show. Appreciate you guys coming on with us.
>> Welcome, welcome. Let's bring bring the energy. It's a pleasure to have you guys. Uh we're going to get right into it. I know I've been monologuing, but uh it's a pleasure to have you guys. I think some of you guys, you know, first time on the show. Uh but uh you know I'll kick around some questions. Feel free. We're just going to have have fun, have a good convo between all of us. So feel free to jump in. Um you know, ultimately we just kind of got a good overview of uh near.com and how uh Matt Henderson at Near is building, you know, this this all-encompassing uh app and they've been able to include several different components. Near Intense is a big part of that. uh you know$23 billion dollars in volume but there's AI components there's wallets there's per there's payments there's intents there's all these things um the change shouldn't matter especially when the agents are the ones transacting I'm curious how you guys are thinking about agents it kind of feels like we go through these like es and flows right where oh my goodness X42 is the greatest thing since sliced bread we need to get in this thing right right now and then it's like okay well you know this thing is going to take a little while to come to fruition and you know we get new updates in in frontier models, you know, then we're going to sort of see that, you know, continuously see, but still it's humans that are making the predominant, you know, bulk of transactions on chain. Um, you know, maybe we'll get into we'll get into that. Where do you guys stand on this?
Do you think we should be building for agents right now? How much time do we have until ultimately this is inevitable? Is it inevitable? Um, Felix, maybe we'll start with you and we'll go down the line, but guys, feel free to just jump in. Felix, I I feel like I got a call on someone. I just want to get get us all get us all uh chipping in here. So, how you doing, man? And uh what do you think about uh agents in this day and age?
>> Um I'm doing great. Thanks for having me. Um I mean, my personal take on agents is if you as you said, if you see if you see the usage right now, it's probably um still very small in in today's kind of importance when we think about what are the things we uh need to improve in terms of experience this quarter, next quarter. um they're the human and and um you know possibly like the next wave of people that will come on chain are probably the more uh dominant target and and and what we're trying to to optimize for. Uh that being said, I think it's very clear that over time more and more tasks will be outsourced to agents and you know while you might do your food delivery or your um stock picking or or other kind of use cases through an agent then as that grows it becomes very clear that that the agent will be the one that also decides on which venue they will potentially execute their trade or their swap or you know whatever they need to do um to to fulfill your intent. to me like kind of what the user the prompt that the user gives to an agent in itself is an intent and and then and agents could you know in a way also be seen as competitors to the existing I guess dex aggregators or intent fulfiller solvers. Um and so we need to ask ourselves what is our uh what is our value ad in an in a world where routing for instance is completely um commoditized and and any agent can just find the optimal route by just being smart um doing that. So, so I think that's that's where we're thinking in like the next, you know, I'd say six to 12 months. Again, I think we still have a bit of time, but but I think that that definitely questions a bunch of our business models and and kind of how we position ourselves. So, >> yeah, the these things are accelerating 6 to 12 months, you know, a lot can happen in that time frame. Um, you know, so honing in a little bit, do you guys think that they're really, you know, do these agents care whatsoever what chain they're on? You know, I don't know why, but the chain you guys remember blast blast like if there's be for some reason if there's the best liquidity for some, you know, long tail, you know, third party third party like swamp. If for some reason the best liquidity, best price execution's on blast and, you know, my my agent can go find that route. You think they're going to do that or or are they like, "Hey, even having some money on Blast for, you know, five minutes to do this swap is is too big of a risk." Like do you guys think that there's there's really any preferences when they come when these agents are making transactions when it comes to chains? What do you guys think?
Is it is it just best price wins?
>> Yeah, I mean I think for the most part um there's a couple things that matter when it comes to uh like what chains agents will want to execute on. And I think even for users, they don't really care so much what chains they're actually on. It's more so like what enables them to do what they actually want to do. Whether that's like buying certain assets that's buying maybe they have a preference for native assets rather than wrapped assets which are you know have additional like custody assumptions. Um I think that um you know once you're just already interacting on chain with these type of assets it doesn't actually matter that much moving from one chain uh to a different chain.
So I think that they'll probably have like preferred um like execution environment where they can just say like hey like I prefer to you know interact on base chain or salana or just hyperlquid or something but you know further than that it's really about like what they can get and just reaching their goal. Like I I think like a lot of the um chain abstraction stuff comes down to um the chain is just friction to get to the actual product that the user wants to get to and uh it just removing all that friction and making that as easy as possible to actually get to the end goal which is not like hey I want to interact on this chain. It's like I want to do this thing like I want to you know pick this uh like I want to work on this prediction market. I want to make this bet like I I want this leverage position. I want this spot position. Uh I want to make this loan. Uh so the actual chain itself I think matters less than um than the actual like intent of the user like what they actually want to accomplish on that chain.
>> Yeah. So almost entirely outcome driven.
Um and you know given that right now you know we may have our assets on a particular chain it presents this ease of use to just execute that that transaction on that same chain. But when the use is still just telling our agent they are able to accomplish this outcome in you know something that's easy for us as humans might be different than what's easy for the agent. could be easier in terms of price or friction for them to go execute on a different chain because it's all just routing through a particular particular mechanism. Uh Hugo, anything to add here as we think about, you know, what kind of preferences an agent might have when they're making some of these transactions on our behalf?
>> No, but it's it's true like you sum up quite well, Tyler, but again [snorts] like it's still at the end of the day like we are human and we're programming the the agent, right? So we are still telling him like where to go and I mostly say like most of the most of the time where you want to go where there is liquidity where where where there is reliability and and some it's like you you still have a preference like the users still will say right you just go on Ethereum because this where all the liquidity is. So well that's >> but this is also the the question on who carries the risk or who who guarantees um certain things like what happens if you actually go on a ghost chain that is has a centralized sequencer and that sequencer goes down while you're trying to execute your trade like you know if your agent does that then on your behalf I think in practice you'll you'll probably still have to like you know tell him that you are fine with taking you know which chains you are comfortable with but if you can actually outsource this and an intent event based system to a solver for instance and the solver just delivers the outcome on whatever chain you need it like let's say you want to yeah swap some tokens and liquidity is better on another chain a solver can possibly take that execution risk deliver tokens on the chain that you want them and then really you can just express things as like you know I want I want to sell this and I want to get as much of the the buy token as possible um to to not have this I guess two-dimensional um preference set where it's like price and I mean I don't want to lose my Or I think when it comes specifically when it comes to bridging another dimension that's very important is time as well certain chains if you use the native bridge you have like perfect or like pretty good security assumptions and um but you might have to wait a couple of days before you get your proceeds whereas if you use some other systems you might have second fast execution but who can you call in case your your your bridging transaction gets stuck and you don't get your tokens delivered on on on the other side. So, I think these are like not super trivial um trade-offs and and if an agent acts on behalf of a human, it's still the human that somehow needs to tell it what what risk um it is is comfortable with.
Are >> are you guys building with these parameters in mind now? Like I I think it is very intuitive to say, hey, you know, I want my proceeds to get delivered to me on my say home chain or chains. like say I only want to like keep my balances on, you know, Ethereum mainet and maybe, you know, one or two other chains. But when it comes to execution, maybe I'll I'll broaden the aperture to maybe five to 10 or 15 chains because my assets are only going to go there for a small amount of time, make the transaction, and then they're going to come back to my home chains, right? And I can set my preference to, hey, I want to maximize speed. I don't really care about liquidity or you know take a little bit longer but find me the absolute best price execution so that I can get the most amount of coins proceeds for my transaction. Are you guys building with this? You know, we've got we've got Kyber swap, we've got cow swap, we've got near intense. Like how much of this is already in practice versus a theoretical design choice that'll get implemented over the next six to 12 months?
Yeah, I think I mean we we touched on this when we'd um delivered swap and bridge which we delivered in um collaboration also with near intense um and there specifically we for now um take some you know take some decisions on like what is the trade-off between uh speed and security like I mean for instance we we trust the near validator set and therefore say like that near is you know as secure as potentially a native bridging solution or CCTP [clears throat] although they have very different trust assumptions.
Um but I think when we when we think about like the ideal world, we would love to outsource this risk to solvers and we would it's a bit more difficult in the in the um well in the crosschain sense. I guess it's it's slightly more difficult than saying I'm on chain A and I want to make a trade and best liquidity lives on chain B. there really we would like the solvers to tell the user hey it might take a little bit longer like I might not be able to to settle you in the next two blocks but um I can settle you in maybe 10 blocks or 20 blocks because I need to go on another chain and source liquidity there and if the user is okay with you know things taking a bit longer but getting a better price then um they can sign off on that and and and we can deliver um we can deliver the experience but we learned also from user early user feedback user testing that the more you try to make a user choose the more confusing and difficult it becomes Um so you know just saying I want to sell and give me as much as possible is a very very easy intent to express. Now having multiple dimensions to kind of trade off a lot of users don't even know what is a sensible trade-off and would like to defer that decision to somebody else but then you know whoever does that decision then probably also you know carries some risk and and responsibility that they might not want to um take.
>> Yeah. I mean, I think just going off of what Felix said, I think it's definitely on the responsibility of the apps and in that case like whoever is developing the agent or like the the harness around that to determine like what the best trade-offs are cuz it's all trade-offs when you can say like oh I want to do everything on one chain, but then I'm uh you know whose custody are these assets in like these wrapped assets? Who's the issuer of these assets? Um what bridges am I using? What kind of execution environment am I on? like what's the trade-off versus going to another chain and doing that. So, it's really up to the team that's actually just behind it to do their diligence and say like, okay, these are the trade-offs that we make and this makes sense because um you know, it puts the u puts the user in the best spot where there's no like no risk of losing funds or you know, running into like you know, poor execution from just no liquidity or like MEV or or something like that. So I think it's definitely on the teams that are just putting these products in the hands of users to say like you know these are the these are the bridges that are going to you know have the most reliable execution and these are the products that the that there's demand for and then over time that all aggregates together and then you have this big suite of products and that that's kind of shaking out right now I think with just some of the um you know the the biggest crypto apps like they're they're kind of getting integrated everywhere like you know everyone's plugging in like Hyperlid for parts because there's a lot of liquidity there and it's you know fairly easy to access through the bridges that we have. Um, so and there there's other purpose platforms as well, but you have to make a trade-off like do I want to go to this chain, do I want to go to that chain, where can I get the best execution. So it's really up to the teams to decide like where they want to put their users and there's a lot of trust involved there. Um, especially like in this space.
>> It feels like I I like what you said. I mean the the biggest ones are starting to get, you know, integrated everywhere and become ubiquitous. like you mentioned hyperlquid builder codes.
These things are are you know underpinning a lot of the per trading that's taking place in wallets I think on near.com as well is a is another good example. The other one that comes to mind is Morpho absolutely crushing it.
Morphos's got embedded you know into Coinbase and it almost feels like it's you know you guys remember like the old DeFi mullet right where it was like you know uh the the modular mullet where it was like sort of these unified front ends and you know more modular in the back end. It feels like now we've got this Tradfi in the front end, DeFi in the back end where, you know, you can have like, you know, uh, PayPal, uh, centralized exchange, you know, these massive brands that have been around for a long time, these financial institutional incumbents that own the relationship with the end customer.
Those are the ones that the customers already trust. The the co consumer behavior is already uh, habitual.
they're already going back to these apps time and time again. Those consumers probably won't change their behavior, but it's on the the impetence is on the applications to implement, you know, the more efficient the new features, those things in the back end and plug into all the things that we've we've got built now. Uh and and it feels like that is sort of the the direction that we're headed. Um I want to kind of get one level deeper in the stack. Felix, you mentioned something about, you know, the solvers if if you know, all of the the order flow routing is going through these LLMs, the the solvers almost get commoditized. And so given that it feels like crosschain, we've sort of got to this point where, you know, intense is the right architecture. Um, maybe there's debate about that. I'm curious what you guys think if, you know, we've kind of arrived at, you know, what we've found is the the most optimal crosschain architecture. Is it intense? Is it something else? But given that intense relies on solvers and and these sorts of things and that becomes commoditized like where is the moat? Where is the you know the juice? Is it in the liquidity providers you know as people and and teams that are building these crosschain intent platforms?
How do you maintain a you know a profitable business if you see parts of your stack start to get commoditized?
Yeah, I think I mean for CO protocol the answer is is pretty clear. It's in the network effects that comes from coincidence of once. So basically users that are trading at the same time, they get batched together into the same kind of execution problem and then they act almost as liquidity providers for one another. And so um even if the optimal route happens to you know involve 10 different pools on chain if there's a user that is either um doing a trade directly in the opposite direction of your trade or even just taking a part of that route and you know has a resting limit order or is trying to make a trade at the same time then we can match you guys together um as a coincidence of want and we can do that at the midpoint so we don't have to pay any spread any uh fees um and so I think batch batching batching batched intense system really have these positive networks effect that the more people use them the higher the um efficiency gain gains are and we are very bullish on um agentized economies to to actually significantly up the amount of velocity and number of trades that are going to happen roughly at the same time because if you look at just the likelihood of a coincidence of what happened today with the current interest the current retail usage on Ethereum or or other chains then you see that they happen fairly infrequently but if to consider that maybe in the future people will have their entire you know portfolio strategies with complex uh trading uh strategies such as buy low sell I mean implementing a lot of trades per day rather than you literally having to click on every trade reviewing the transaction signing on it. If you can actually give your trading strategy to an agent that will trade on your behalf, then we think that that amount of of orders will will will go through the roof. And so um any system that can can have efficiency gains from batching people together will still retain um mode and and and thus be the natural choice even for an omni intelligent agent to to place their orders through.
>> Hugo, uh Tyler, anything to add?
>> Yeah. Yeah, sure. Uh so it this is true.
So actually I have another take on that because uh we we speak a lot about near like a lot of like for transaction for swap you also need atomic transaction like it has to be it has to be symbiotic because you know most of the time a lot of cost of solvers rely on our infrastructure kber power maybe hundred of million of volume per day and some of them are solvers that they just don't have the inventory they just add on the liquidity or some of the liquidity and they really need like instant transaction and so yes of course they will they will go where the the execution is the best and and the and this is and this is just like via DEX via liquidity pool so so no no it's a it's a really symbiosis relationship between both and and and you need and you will always need both in the future.
Yeah. And one one more take um which might be interesting. Uh you asked about the you know solvers versus and intense protocols versus like AMMs. Uh you know working on AMMs for a while. I worked on Thor chain for four years before uh joining the near intense team. And just like my take on uh you know AMMs versus um intents and solvers um like AMMs are really good because they're always on but they're extremely capital inefficient where it's the complete opposite for um intents and for solvers where you have this extreme capital efficiency because all you need is a market maker to come on and say yeah I'm going to provide liquidity for this one asset and then they can just be that liquidity provider with a pretty thin book. they can make trades in either direction which is really nice but um you know you see like you know on like 1010 last year where you know market makers step back and that becomes a problem when there's no no liquidity available when you know they decide I don't want to provide liquidity in this moment. So it like on on the norm it provides much greater capital efficiency and just a better experience for for users. we can um onboard assets much easier um just like list more assets have really deep liquidity but um in those edge cases they that's where the AMM really shines but then you have all this capital that's just sitting there for um for a long period of time >> um not you know not really doing much impermanent loss it's very hard to get deep deep books in a in an AMM so there's definitely there's trade-offs uh either way but I think you know in terms of where this like space is going um you know I think that intense is definitely just the the more practical way of of going about it. And like, you know, I I think, you know, it's also, you know, keep coming up more and more that like TVL is kind of a liability where that's just like value that's kind of sitting there that is, you know, could be, you know, exposed to like some kind of exploit or or something. Um, so capital efficiency is definitely like a big gain in uh in a couple different uh ways, especially when you consider that.
>> Yeah. And and so um give could Tyler maybe you could give us a sense just starting from the near.com near intense you know vantage point you know we've got kyers swap we've got cow swap and then obviously near intense itself could you give us some insight into how you guys all work together like you know who plugs into who who's the front end who's the back end like could you guys just kind of take us through like what the life cycle is from you know from that vantage vantage point. Yeah, I mean we kind of do it all like where we have like this so on the back there's this network right there's there's the near L1 and then on top of that there's our solver network which has liquidity on you know on Bitcoin Zcash Salana Hyperliquid and then that all connects on the near L1 and then we have um our one-click API product which is like it sits on top of the intense architecture that's where people come in and say Hey, I want a quote for this. I want to do this trade. I want to do that trade. And then that API product we plug into everywhere. So I think that's what's plugged into cow swap right now. The integration we did with swap and bridge.
Um and that's that's mostly how we work with a lot of different projects is we have liquidity on your intents and then someone says hey we want to you know access like your bridging product and uh being able to swap crosschain like uh you know trade assets and they plug in our um our APIs. they access uh they they have access to our solver network in in that sense um and then you know people say like oh I want to provide liquidity they come in as solvers um and then near.com is another implementation of our oneclick API where that's a like you know something that is like internal to like our team and um that's just like our implementation of it like what we think is just a good user experience and something that we want to you know just just grow internally but mostly we work with um you know external teams just shipping out like our API product, but it's all built on the same liquidity base. So that's, you know, liquidity begets liquidity. So um the more people get plugged into that, like the stronger the book gets and the better execution gets and just good for everyone all around.
And then as as Hugo said, I guess on the Kos website side, when you're when you're trading, then your order gets auctioned off also with a network of solvers. Um, and they're I think Kyver I not they're not called Kyros but like a Kyos affiliated team runs their own solver and then some some other solvers might tap into any any API including Kyberox one to to to find the best route. Um, >> got it. Yeah. And so you guys all sort of have uh like liquidity and inventory on your own, right? Because you've got solvers that plug in, but then you also provide APIs out. So, you know, a solver uh is sort of an independent, but they can be solving for both near intense and cow swap at the same time. Uh if they're, you know, solving over on near intense, but the order comes in on cow swap, you know, maybe you guys route that over to near intense and vice versa. Like order might come in at cow swap and you know, uh you guys might um you know, plug into the solver uh from the near intense side. Kyber swap, same thing. Orders can come in, but you guys can also execute orders. guys all sort of do you know this re this consumerf facing order flow routing but also have a network of solvers underneath the hood and also run your own solvers is that is that the case at least I would say cow protocol doesn't run its own solvers we've been for a while like contemplating if we should but we think that like in order to make a neutral and um yeah like to to basically be an good arbitr arbitrator of the competition we don't want participate in it. It might change at some point, but at least right now we're not running solver ourselves >> and and so yeah and yeah, Hugo if you had anything.
>> No, no, no. I just I I wanted to mention that is also not our case. We're not running our own solvers because we want to be like independent and so yeah, we let the the third party do that for us.
>> Yeah. And so, you know, given, you know, order flow, routers, all these things, you know, especially as more orders tick up, you know, Felix, you mentioned coincidence of wants is sort of where you guys see a lot of, uh, you know, opportunity given with, you know, the the ri the potential rise in orders as a result of agentic transactions. Hugo uh, and and Tyler, where do you guys see the moat? Like where do you see the most opportunity in the intent, you know, solver stack? Do you think it's at the routing point of view? Do you think it's at the solver perspective? Where do you think that there's still a lot of opportunity here?
>> Yeah, I think that um the opportunity is just in just getting more it's liquidity, right? Like it's how good can you like what assets do you have um and then how good liquidity is there for that? But I think the real mode is just distribution. So like I mean I think that's been discussed on you know on tech Twitter for you know for forever now but you know the real moat is how many users can you get it into the hands of and it does that provide some kind of value to them does it help them make the trades that they want does it help them like you know reach the goal that they want to um so I think that the moat is just figuring out like what products that people want whether that's like equities outside the US through like RWAS whether that's like exposure to like you know things like gold like XA UT um you know from tether gold um or like another you know what what exactly does that user want and then just providing you know a good user experience and good liquidity for that um I think you know whoever solves the distri like whoever gets distribution like wins the game not that you know there's like an end end to this but um like whoever actually gets that in the hands of users and like at a at a mass scale is um you know whoever benefits the most and um you know just makes the most difference to people. It's like providing a good service that that users want. So um yeah, I think that's that's the name of the game is just figuring out like what what do people actually want to do on chain? What makes people's lives better? Um and then how do we accomplish that?
And si similar question but I guess a little bit a little bit more um you know quantifiable you know and the near intense uh you know place is where I I I have the most insight into the numbers at least at at this point. Um you know $23 billion and $40 million of fees.
Does that seem like is that pretty much on par with what the ecosystem is pricing this at just generally? Like I know these are near intense numbers, but do you guys as a group think that this is like are we charging too much? Are we charging too little? Because it seems like, you know, if we're on the scale of billions of dollars, it almost seems like, you know, that you could be charging more and there could be more fees at play here. But you guys are the the experts when it comes to, you know, maybe pricing models and and the right monetization strategy here. what is sort of the baseline for um you know how much it costs to to conduct one of these transactions? Um I'm not sure exactly how much that is in terms of basis points, but $40 million on $23 billion of volume. Does that feel like it's on par with the rec rest of the ecosystem?
And generally as a group, do you guys think we're moving in the direction of like more monetization because we're providing valuable service or you know we're going to bring this down as you know more of this technology becomes better uh and more efficient.
>> So I think monetization in the space is an extremely interesting question and problem and we had our fair share of it at at C Protocol last year when we started charging volume fees. So if I get the numbers correctly, it seems about 20 basis points on near intense, which is I guess less than other um apps that control a lot of price insensitive users like MetaMask and and Phantom or or some other um you know big big retail focused apps. I think when it comes to specifically Ethereum whales and and hardcore users that is a ton like I think Kazop we tried to where we started charging a two basis point um volume fee last year and we saw um a big change in uh in in user behavior basically people stopping to use cow swap for this two basis point fee even though we think that our execution on average is at least two basis points better than what they would get elsewhere so in the sense that we you know you'd still best off to go to go but okay I mean I have to say that of course um but I think there's there's other ways that we see competitors for instance taking fees that are less clear to the user um so 20 basis points is somewhere in the middle I would say um there's there's different ways of how you can take fees one way is to just um basically extract positive slippage right like if if you get a better outcome than what the user expected you're not forced I mean there's no no law that tells you you have to give the user like what is currently available on on chain so I think a lot of competitors um just just capture that positive slippage entirely and say like as long as the user gets um the user cannot get more than what they basically signed off at. And the other thing we we see is sometimes just on the quoting side. So you don't actually call it a volume based fee. You just quote the user slightly worse. So um I think this has also been done on Revolute for instance when they say like your your foreign for forex trading doesn't have any fees but then the the rate that you get is actually not that good. So in in a sense that's also a fee. Um, and I would say like as them as the industry matures and and especially like on the DEX layer and like the the the close to the actual chain layer, uh, fees are being competed away and it's it's very hard to to take to take uh good fees where you're still able to take very nice fees is if you own the distribution and you have retail users that are maybe less price sensitive. Um, so again, that probably is one of the things that that um you should look out for when when you're trying to acquire users.
Hugo, Tyler. Yeah, please.
>> Yeah, I think that the it depends on what product that like someone's using.
So, I think um like for example, if you're doing a stable coin to stable coin swap, you're probably pretty price sensitive to that. Like if you put in, you know, a,000 USDC, you probably want pretty close to a,000 USDT out on the other end. Well, I think um like where most of the monetization comes in is from uh you know unstable assets that Bitcoin to Ethereum or Zcash to Salana or something. Um I think that there's much more margin in there and uh there's just you know there's much more of a market like to go back and forth and there there's not like a book that's like you know a Zcash to Salana book that um is going to have like super deep liquidity. So um I think that there's just like a lot more space in there to actually like monetize uh swaps. Um but you look at like the hyperlquid when when people are you know trying to get like you know high gains by you know playing with leverage and per um they don't care so much about like you know you you could charge them you know however much fees and it really doesn't matter because you know that's just like that's pretty small compared to like what the like the riskreward is on what they're trying to do. So it really depends on like what the action is. So if you're if you're going stable to stable, there's, you know, basically no margin, maybe one basis point, two basis points if that. Um if it's like some kind of like, you know, high-risisk, high reward thing, then there's probably higher fees involved in that just because um that's just part of that user's um just uh you know, risk tolerance for for doing that. Um so it it totally depends on like what the product is. I think for for yield in this space, um it it it looks different as well. So, um I think every single um like use case and and way to use swaps um has like some kind of different pricing model with it. I think that I mean I would like to see you know crosschain swaps as cheap as possible because um I think that's really how you you know get people to uh and especially institutions and businesses to come on chain and adopt it. like they're not going to come uh and and use onchain products if they're more expensive than a traditional product and they offer less um less guarantees and and and safety um and you know other headaches that might come with moving onchain. So um yeah, I think it totally just depends on what what the user is trying to do.
You know, Tyler, intuitively, I also want to see uh crosschain swaps and transactions be as cheap as possible, but then I'm like, wait, we're in this like very revenue oriented market and all of the fees that are generated via crosschain swaps get, you know, are are get used to buy back the tokens that we're holding. So it's like yeah I want my fees I want my fees to be less but I also want the chains or or I want the you know the the companies to have revenue so that you know these these tokens go up and and you know these things do well. So I I think that you're right. The long the long you know term view is let's bring these things down.
The pie gets bigger. You know lower transaction fees means more users.
Companies still you know do quite well when we have a large swath of of user activity. You know all paying a little bit of fees still generates a lot of revenue. Companies tokens still do quite well.
Guys I think that's it. Appreciate your time. Um thanks so much for coming on.
It was a it was a pleasure to uh to go deep into the intentbased crosschain swap uh business uh industry here. Um and uh appreciate you guys joining us on virtual near day. We will uh we will keep up with uh everything happening the intent uh the intent market. So uh hope to have you guys back again soon.
>> Thanks for having us.
>> Awesome.
>> Thanks gentlemen. We'll see you again soon. Cheers.
All right. Yeah, it's a it's a funky uh it's a funky mindset to uh to think here, you know, uh you know, we want to bring fees down and and the best, you know, example of this right now is Ethereum. It's like Ethereum is doing everything in its [snorts] in everything in its power to bring transaction costs down, right? Really trying to commoditize Blockspace, saying like it is so cheap to make transactions here.
um you know all of those those fees come down. It becomes extremely easy to conduct transactions, extremely affordable to conduct transactions. But then you get into this revenue crisis.
You get into this place where yeah fees are extremely you know extremely affordable yet the company that you want the token or the equity of the company to go up that thing isn't generating a lot of fees because they worked so hard on you know making these technological improvements to bring the fees down. So the solution to all of this is expand the pie user adoption. We need, you know, a large swath of users in order to uh overcome and make up uh these these much lower fees, right? Because uh ultimately the commulative, you know, volume and activity here is what ends up uh increasing uh the pie for everyone.
And you can get low fees and uh revenue that helps these tokens go up. I can see our next group of panelists uh in the back. We're going to bring them up on the stage now. And the next panel is all about confidentiality. Uh we've got Dylan from Near, we've got Daniel from Zod, we've got Joel from Dash, and we've got Maya from Starknet. Uh everyone here is doing uh everything in their power to make sure that we maintain the cipher punk privacy ethos that this space was built on. Uh we've got, you know, big panel here. Uh so we're going to go, we're going to have some fun, guys. you know, I might just ask some general questions and then, you know, feel free to just jump in. Uh, but first, it's a pleasure to have you all here. Welcome to the show.
>> Thanks for having me. Yeah, it's great.
>> Good to be here.
>> Awesome. Great. Great to play be here, guys.
>> Absolutely. Thanks so much for joining.
Um, you know, privacy and crypto has uh evolved through multiple layers. You know, started off with Bitcoin. We realized, you know, chain analysis helped us realize Bitcoin wasn't so private. And then uh you know we ended up in a in a world where uh a lot of teams now are making this a a forefront and a priority. Uh we're implementing it at different levels of the stack. The asset layer, the contract layer, the smart contract layer. Um you know a mix of ZK proofs uh are in there as well.
TEES are in there as well. Guys, what I really want to talk about is how you know we're thinking about privacy in an agentic world. Um, it feels like the inevitable trajectory of this space is more agents transacting. The market share of agents is going to increase relative to humans on chain. How should we think about privacy in an agentic first onchain economy? Open question.
Feel free to jump in.
>> Yeah. So I don't have a lot to say on this but one thing I will say is in this increasingly agentic world uh what we have to understand is agentic um analysis is what's going to be also u on the rise and so with for example just countless onchain transactions for everyone to watch pattern recognition where flows are coming in and out of wear especially in the age of exploits etc. There's going to be a microscope on this right now. And so I think that before, like for example, my first Bitcoin wallet ever, I just had one address. And when they upgraded to HD and I got a new address every time, I was that's showing how old I am. Uh it's kind of annoyed me. I'm like, why? I wanted my favorite one. I didn't understand why privacy is that important. And I think a lot of people hid because no one was looking. But now everyone is looking. So you need to protect yourself.
>> Yeah. One one thing I'll add is a a agents don't care guys. They're not like humans. They have no allegiances. They want to use the best tech. Um and so when you look at Agentic Commerce, especially on on chain, our general take is um you know, they're going to be using tools, products that allow them to get where they need to be, allow them the most anonymity. Um and so one thing that you know we're prioritizing in Zcash is the orchard pool, soon to be Ironwood, soon to be Tachion. um the higher we get that TBL and encrypted money, more money is going to flow in, right? Liquidity begets liquidity. So my general take around the agentic economy when it comes to privacy is agents don't care. They want to use the tools that allow them to be um the best versions of themselves. They've got no allegiances to Bitcoin, to Zcash, to Near, to Starknet. Um, and you know, I think if we can prioritize, um, you know, the shielded pool, um, that's going to be one thing that that draws them in.
>> Yeah. One thing that I, you know, I I'll just piggyback off of what Dan said, I think, is that, you know, agents don't care about which assets they're using, right? So, with Near Intense, we're able to bring this privacy set that a lot of people are focused on. We're able to bring this to our 140 plus assets across 35 different chains. And so we enable the most flexibility for these agents to be able to to transact and do these e-commerce. And this is one of the ways that we feel from the near side we're well set up for an agentic future.
>> So I would say that one of the most important things for agentic payments is scale. So you need both privacy and scale and you also need a setup protocol that enables privacy and scale even if the agent is not even malicious but manipulated. So you know you need a protocol that supports real privacy, enables real privacy, users that are responsible with agents and also you need scale.
>> Yeah. Um I think there's there and I want to get into you know how some of these uh these privacy enabled features can be combined at different levels of the stack because I can I can even imagine you know like uh let's say we have you know privacy at the asset layer right so that you know just transfers are encrypted and balances are encrypted obviously we want to do more with our assets and so you know getting this at the smart contract layer at the application layer all of this is going to be important so that cuz even if if you let the cat out of the bag just one time, you know, all of a sudden all of your data is exposed. You need end to end privacy for this thing to really work. Um I I want to start just with the near zcash, you know, dynamic because uh well it is virtual near day kind of in you know I want to talk about this one use case because I think this is the case but if it's the case it feels really powerful and I haven't seen a lot enough people talking about it. You know we know about this encrypted pool on Zcash. you can put in, you can shield your assets. You know, obviously the idea here is that you want to do, you know, more with your assets, right, Daniel? So, like it's not just enough that we want to have this shielded. We want to be able to stake it, trade it, swap it, move it across all these different applications and for it to remain confidential. Maybe that's where near intense comes in. Or maybe there's different smart contract privacy features that can come in. Can you guys just talk about, you know, let's say I have Zcash that I've chosen to shield.
Is it possible to then leverage something like a near intense to, you know, swap it or earn some yield on it somewhere and keep that data confidential and shielded?
>> Yeah, Dylan, I'll start and then I would love to toss it over to you. Um, look, within you can ask, there are many different folks working on Zcash.
Everyone has differing opinions. Within Zado, our priority is to make Zcash usable. Um, and you know, we're kind of dead set around this idea of encrypted money at scale. Uh, so our focus is making Zcash within the Orchard pool usable and allowing people to get in there, get out there, um, and actually spend it. So, Near has been an invaluable partner as has Salana um, with with them listing Omnisc, which which Near was heavily involved in. You can now buy Zcash, you know, on Hyperlid spot as well. Um, so what we're seeing is people are willing to buy Zcash outside the ecosystem. Uh, but we want to bring them back into the orchard pool. So my focus and my goal is to make Zcash as an asset usable. Uh, so what does that mean? Spending it in the real world, paying with friends, everything that you would think of from like a Venmo or a cash app, we want to enable with Zcash. So even if you're profit taking Zcash into into USDC, we want to keep you within the Zodel app and within the ecosystem to some extent. Uh so to that to that lens we will be leveraging near quite extensively for swapping as we already have. There are many other venues now looking to integrate Zcash.
Um and you know we are extremely grateful for the NER team to to take a bet on Zcash as an asset and bet on this team because they were one of the first to allow for us to make it ubiquitous.
Uh but now it's mostly getting it in the hands of users and allowing people to actually pay for it um using using encrypted cash.
Yeah.
>> Yeah. Absolutely. I' I'd say from the near side, we've definitely enjoyed the partnership with Zcash. It's been one of our, you know, more successful integrations and and ecosystem partnerships. The way that we look at it is we have two different kind of methods of our go to market strategy, right? One of those is on the B2B side. So on the B2B side, we enable swapping in and out of Zcash into any other asset as as Dan mentioned, you know, making it more usable and, you know, whether you're taking profits or whether you're entering into the Zcash ecosystem, making it easier for people and users to do that. The other way, and I I think you might have heard earlier from from Matt from the Near.com team, is we have the BTOC kind of app with Near.com. And with Near.com, what you can do is you can hold all of your assets in the confidential balance and other things that we're enabling soon. You know, you'll be able to enter your per positions confidentially through hyperlquid builder codes. You'll be you'll be able to stake and earn confidentially. We've got earn vaults that are, you know, just on the horizon that are coming out and, you know, working on things like private send and these types of things where you're allowed to take all of your assets, you know, especially Zcash or Dash or USDC or what have you, and you're able to do anything within the realm of crypto and web 3, uh, in a confidential and private way.
Yeah, we have someone in the chat here that says uh personally I would love a way in which AI agent can start accepting near as payment for subscription so it can drive near token utility. can imagine uh you know uh we're going to see agents start to do all kinds of crazy things and accept all kinds of crazy payments you know uh near Zcash Dash Stark all of these things you know these these are AI agent money um Maya and and Joel I also you know I'm I'm curious how you guys think about privacy at Starknet and Dash you know where do you see yourselves implementing um the privacy elements is it at the asset level is it at the smart contract level or the chain level and how do you think through some of these different mechanisms?
>> Yeah, so at Starknet we implemented what we're calling a protocol level privacy pool. It's an canonical privacy pool and the way it works is anyone can shield any asset in the same pool. So it's the same anonymity set for all token types.
Um we have client side proving since start is zk rollup. We [clears throat] have the same infra and language and prover on client side and on the server blockchain side. So everything is compatible which enables us to have very rich a very rich privacy ecosystem. So we can have um basically any multiol operation happen privately or if it goes outside the pool anonymously. So you can today uh use uh any app on starknet anonymously via this pool. H all you need is to write a helper or an anonymizer contract for the app. H and additionally we're now having integrations with the other blockchains.
So we're building a product called off market which is basically a anonymizer for poly market which means you can trade on poly market but uh it makes your trades unlink unlinkable. So we just announced this so and it works pretty fast compatible with stocknet infra and with EVM wallets so seamless.
Yeah. Maya, did you see there was uh just today there was someone that they do the teleprompter, you know, like when the president reads off the screen, they have the teleprompter. The person who was in charge of the teleprompter was betting I think on call she maybe poly market you know what the person would say.
>> Oh yeah, that that happens a lot. And we actually because um we worked on privatizing stuff on poly market. We actually have a very good tracking of what happens on poly market and I have a detector for insider trading on poly market which works pretty well. So [laughter] I find a lot of of positives that disappear the the next day from the website. So so it's >> nuts.
>> Yeah.
>> And and and Joel on on the Dash side, where are you guys thinking about privacy, you know, protocol level, asset level, how do you think about this?
Yeah, definitely protocol level and after all we are the oldest I would say surviving cryptocurrency that had any explicit privacy function built in. Um, so this this might be a little bit of an alpha because we haven't put out the announcement yet for another couple days, but Dash has integrated a shielded pool based on Zcash's orchard technology and it is live on mainet currently right now and we just have to launch the the p the public beta of the wallets that actually support it but it's live on the the network. So, uh, we're adopting the same shielded technology as Zcash. We have a few different hopefully optimizations that make things work very fast like 1second confirmations as well as I think we can sync an entire wallet from zero in about 20 or so seconds. And syncing from syncing wallets has been the bane of any privacy coin sort of existence. And we're also adding um a shielded asset pool for our rich token functionality. So you could for example uh once it's launched do shielded stable coin transfers with 1 second finality which sound you know in my opinion pretty cool and uh we're very very happy that Zcash has paved the way with so much of this and there's um I kind of don't I like that there's so many approaches to solving the the fundamental privacy issues in the space and I like that we have different but similar sorts of paths that we can kind of AB test things. So, I know that for example, shielded assets is something that GCash basically came up with and it's not 100% clear from my understanding if that's going to be implemented, but we're more than happy to test it out and make the mistakes and show if it works or not and see if there's a product market fit and then grow the whole space together there.
>> Uh, anything to add from the Zcash side?
Uh shielded.
>> Yeah, [laughter] >> my mic was shielded. Can you guys hear me now?
>> Yeah.
>> Yeah. Listen, I mean our or I guess our take internally at Zotle is as I mentioned previously, there are many different folks working on Zcash, many different organizations. Um our take is to focus on making Zcash a top three, top five, hopefully top three asset. um at at like a shielded level and bring the usability around that to make it you know en enable like the promise of encrypted cache and once we get to that stage we can maybe start exploring you know ZSAs or Zcash shielded assets um but until then as I mentioned previously we're going to be relying on teams like Near on teams like Hyperlquid on teams like Salana um for that kind of network effect of usability outside of the shielded pool whether that's loans whether that's AMM M yield whether that's delta neutral strategies. There's a lot of design space we can now do on chain because of ecosystems like near salon and hyperlquid. So that's what we're going to tap into. Um but for zodle we're going to be focused on user experiences. We're going to make Zcash usable and we're going to get it to a top three asset.
>> There we go. We're going to be we're going to be right there with you. That's right.
>> Um yeah, it's it's huge too. And now that you know you you have the initial foundation and the basis all of these use cases open up um you know you we've got the opportunity to you know you can do uh stable coins backed by these private assets. We can do all all kinds of fun fun DeFi things as a result uh with these mechanisms. Um given that these things are starting to become ubiquitous. We're starting to implement privacy at different levels of the stack. All of this is starting to create this end-to-end encrypted, confidential, shielded ecosystem.
I'd imagine, you guys probably would know better than I would, but regulators starting to pay a little bit closer attention. How do we work with regulators without letting the cat out of the bag? Like there's a and and we had um Alex Chevhenko, you know, the the near intense co-founder on the show yesterday talking about his idea of practical privacy. um he was just you know talking about how they're in near intense I think this is soon to be integrated there will be this compliance council if you will uh which will have this opportunity to create this uh this key which will give uh you know if there's a warrant or something that's justified this council will have to approve it you know he gave the example if it's North Korea that's saying hey we want access to your entire system no the council is not going to approve that but if it's the FBI they come in they say hey we have reason to suspect you know to to reasonable suspicion that you know there's some fraudulent activity or illicit activity happening here. We need you know access or or visibility into this particular account. There is a mechanism to say hey you know for only this account and this account only and for this time frame only and there's different levels of granularity you know there are uh mechanisms in order to ensure compliance.
How do you guys think about this? you know it it we've also heard from you know guys like Vitalic who take privacy very seriously like the walk away test right like even if I wanted to hack or or give some sort of you know visibility underneath the hood of the private elements I couldn't right it's just it's that technically sound open question how should we think about this we want to make sure that we're we're remaining confidential and shielded but we also don't want to get taken down by you know the the state actors at large. What is the right What is the right balance here?
>> Yeah, I I can go ahead and start on that, Robbie, since you you kind of gave the layup and already did the bit a little bit on on the selective disclosures that we have uh with confidential intents, but yeah, you know, we're we're happy to work with law enforcement agencies and partners uh when needed. We look at it as practical privacy where a user doesn't want their activity tracked or linked, right? Like if I'm thinking about something like payments and I go to the bodega down the street and I pay $6 for something, do I want that person to then be able to use some kind of AI tool to figure out my entire balance and all of my history and transactions and what I was doing on pump fun and this and that? You know, no, we do not. At the same time, we don't want to be a a tornado cash or, you know, a mixer or anything like that.
So, we have partners in the space. You know, we have a full compliance team that we've hired up uh in the recent months that has been taking very serious look at this and and being proactive with law enforcement agencies and it's something that uh we definitely are taking seriously. On the flip side, we think that most users uh as long as they have a a decent level of privacy that you know that's okay with them. So on the BTOC side that is the most secure and most private uh setup that we have with Near with Near.com there's no RPCs for you know users to query you can't figure out user balances you know you don't know if during Joel's alpha spill if I just bought some Dash on on near.com um and then on the on the B2B side yeah on the B2B side u you know we have different levels of confidentiality that are are you more practical and on foreign to foreign chain transactions it's a little bit difficult but we're figuring out ways uh to implement harder and harder security measures to make it harder for users to figure out or security researchers and and hobbyists to figure out what's going on. So yeah, we're taking a very much uh practical approach to this and working with law enforcement agencies when needed um while keeping users safe and confidential.
>> Yeah. Yeah, one thing one thing I'd add to that um you know we we are powered by Near Intents. We've been powered by Near Intents for a while. So we kind of they they have their own compliance when it comes to screening wallets into and out of specifically into and out of the shielded pool um using their intents.
But one of the things that you know I was shocked when I joined Zotle and Zcash was the lobbying efforts that have gone in from our ecosystem into communicating to DC um about what we're doing and what we're building. So Paul Brigner on our team, he's our head of policy um is in DC petitioning for Zcash, petitioning for uh encrypted money. Um, and there's this saying within Zcash, within Zotle, um, and within our community that the TAM for freedom is larger than the TAM for criminal activity. And we stand by that and we believe that. And, and Paul is doing amazing work in DC lobbying for, you know, this this little niche called privacy within crypto.
>> Very cool. Yeah, >> I want to add that like we are also we have a lot of compliance built in very near very similar to what you're doing.
Um I do think that it can't be the case that users on the on chain have less privacy than users offchain right so if you have some privacy while paying with your credit card you have some privacy with your bank account it can't be that everything has to be uh out there on the blockchain it has to be legally reasonable to have some level of privacy as long as it is monitored to a reasonable extent and um there is some entity that can answer regulatory inquiries, the law enforce enforcement inquiries. I think that any like anyone that's not actually trying to use these services for criminal intents, then shouldn't have a problem with that.
>> Yeah. Joel, anything to add on the regulatory front?
>> Yes. And this is one of those things where um we kind of have this this struggle between when this technology was created. Let's just start with Bitcoin, right? Uh decentralized financial systems and and now data systems. When this was created, it was openly created with the possibility of being an adversarial technology. Satoshi was notably very anonymous, had no VCs behind him that we know of, no kind of no traces at all.
Basically operated under the assumption that if discovered, he would be arrested and thrown in prison like many of his predecessors were with other forms of ecash out there, digital gold, etc. Uh, and so that's kind of the roots of this technology. That's the roots of the value proposition of this technology is it works for no matter who uses it, no matter where in the world, for whatever.
And then of course we have the the other side of things which is in the real world people live under jurisdictions and under laws and if they can't use something in a legal way some people you know criminals will use it uh protesters will use it desperate people will use it but the bulk of humanity might not. And I believe this technology is for the human race. this will make us all better off and so the struggle also like laws, regulations, social norms etc changed wildly over the years. Um I guess our approach but again Dash is a very decentralized project. So this this you could have slightly different takes around but for the most part the the understanding is that we're building something that enables the most use cases for the most people in most of the world and uh we're trying to make it useful to everyone. So if someone needs to use it for something for some kind of human rights activism where they're they're terrified of their pay payments being discovered by whatever the local warlord might be it works for that if they need to do it in a fully compliant kind of like you know ma compliant AMLR all that EU stuff that we're currently dealing with kind of navigating that situation they can also do it through that um kind of through that thing. So that's one reason why I think that for example privacy is uh absolutely essential and it's also essential that it's not necessarily mandated because people do have either legal reasons or other use case reasons why they might not want privacy in some cases. I have operated some public funds before where I intentionally concentrate all funds into a single published address so the whole world can see this is how much I have this is how much I'm working with I'm not scamming anyone and there's a use case for that but my private funds nobody's business >> have freedom to choose their own freedom tech so that's why we're all building different things right >> yeah and I I think it was Alex last yesterday that talked about how um you know effectively think if we think about cash today, right? We've got you know cash in in a bank in a checking account and then we've got cash on our counter and then we've got cash in a bank vault. like there's different varying levels of convenience and security slashprivacy that we can use and ultimately money is a technology and so it makes sense and and I gave the analogy that you know you don't want to you know if you're trying to you know dig a hole you don't want a hammer you want a shovel like there's different tools for different purposes and and you know therefore uh we should have different flavors of private money in order for you know to to utilize for different purposes Guys, thank you so much for joining. It's all the time that we got today, but um you know, the uh the TAM of privacy and freedom is is absolutely enormous and uh you know, I uh I I really appreciate you guys all coming here today and building freedom tech. Uh it is very very important.
Really appreciate you guys coming on.
Stay confidential, stay shielded. We'll see you again soon. And uh thanks again.
>> Thanks for having us.
>> See you next time. Thanks everyone.
Guys, we are staying private. We're staying confidential and uh we're staying nice nice and liquid, unlevered.
Markets are looking pretty good, pretty chill today. Let's see how the chat's doing. Privacybased stable flatcoin. I like it. I do think as we get more Zcash uh near all these things are going to open up in terms of their DeFi ecosystems, but it's again it's important that we maintain privacy and confidentiality at the base level. And that could mean, you know, when it comes to a a smart contract, it's at the contract level, could be at the chain level, could be at the asset level. If we get it at the asset level, and then we can port that private asset all across the ecosystem and use it in DeFi, I think we're going to get some very, very interesting use cases. Um, I do think a private collateralbased stable coin is going to be extremely valuable. Could see a Zcash stable backed stable coin looking pretty good.
Shadow Hell says, "This is a good project." C Money Music, so the FBI can't be corrupt. Oh boy, that is a whole another conversation. I think there are some streams and podcasts that uh do quite well in in that regard. So Near will be happy to give our data or crypto to the FBI if they simply ask. I don't think it's that simple. Remember, there is this compliance council, right?
They're trying to take this approach of practical privacy. see money music I understand the very you know the libertarian uh viewpoint on this but I do think that there's you know varied ways of approaching this question you know you don't want the the law enforcement to shut down the protocol right you you know then there's no protocol to use for privacy right um the important thing is you know just don't do anything illegal and you'll be fine we don't want to enable illegal activity on the protocol uh flippa thank you for the Zcash alpha now time for the near alpha. I know you guys want to understand the supply constraints on the near token. Uh let's see. Do we have someone that can speak on this? I think maybe towards the end we're going to get uh we're going to get, you know, maybe George or or someone that can speak on the near token. I know that's what you guys want to hear about. Uh but again you know more activity more use cases more volume more revenue for near means more revenue and value acrruel to the token. Uh if we can do that and it can overcome any sort of inflation you know we're we're in a position for some some explosive movements ocoy personally I will love a way in which AI agent can start accepting near as payment for subscriptions so it can drive near token utility. Man, I think you're going to be pleased about what is coming down the pipeline. GMG GMC C Money Music, I got some near. Appreciate you guys rocking with us. Stay tuned.
More alpha, Zcash near, all of the above. We're going to bring up our next panel in just a moment. This next panel is called Cheaper, Smarter, Private by Default. I think this is going to be uh a really fun one. Uh cheaper, smarter, private by default. Private model routing isn't a trade-off anymore. It's how you get the right model for the task at the right cost with the right hardware guarantees underneath the infrastructure making privacy the default instead of the premium. That is what our next panel is going to be all about. And so we're going to bring up uh our panelists in just a moment. This one's going to be fun. We've got guys from sort of outside of the crypto world but coming in on the AI front talking about the privacy element as well. uh from Intel, from Brave, and as well from Near AI. I'm excited. We've already talked a lot about Near Intense. Now, we're going to get a little bit deeper into the AI conversation. Um I see our panelists are already in the back. So, why don't we we bring them up on the stage and we get this panel started.
>> There they are. Guys, welcome uh welcome to the show. It's a pleasure to have you guys.
>> Hey, Ruby.
>> Hey, Robbie.
>> Absolutely. Absolutely. Welcome. Thank you guys for joining. Uh we are about halfway through virtual near day. It's a fun online conference we're having today. So appreciate you guys rocking with us. Um as I said, you know, we talked a lot about some of this, you know, onchain financial elements uh that underpin a lot of these these uh transactions. We've talked about private transactions on chain. Um, but you know, I'm excited to get into uh the AI conversation because I think a large portion of our industry feels as if AI agents are going to be driving a lot of the volume of transactions that we're going to see going forward. Uh, maybe, you know, in a matter of months and especially for years. Um, so again, name of this panel, cheaper, smarter, private by default.
There's really no more trade-off, right?
when it comes to privacy, uh we can effectively get just as good intelligence uh and it can be confidential by default and private so that you know we're not sharing our data uh you know all across the internet. How close are we and and by the way guys I'm just going to ask kind of open questions feel free to jump in and you know we don't need to go one by one you know guys we can just you know we'll have a fun chat here. So question is how close are we to a world where users developers can just they can just automatically get the right model for the task like you know maybe we can start with these general models but then there's like you know these these subm models underneath at the best cost you know strong hardware backed privacy guarantees built in from the start. These things are getting incredibly sophisticated right off the jump. How close are we to just you know getting like these big general sophisticated models and actually getting some you know sophisticated uh you know agent orchestration underneath to to fulfilling these tasks feel and and I just saw news from uh Jensen today that said you know we're still at the beginning of the uh the AI cycle you know I think he's incentivized to say such things but where do you guys where do you guys stand on this uh how far along are we to to really getting a flourishing uh you know uh a gentic world out there.
>> I think we are like really close to be honest like we've seen like especially with these very recent models coming out like JM 5.2 to uh Kim Kimmy 3, Kim K3 just coming out today. Like so all these models can be run inside our like private enclaves we call them and uh this basically enables everybody to use like uh amazing intelligence but coming out of like this uh this like privacy preserving cloud that that we built with different partners and yeah I think it's uh it's coming along. I think now there is like no like the the distance between like open source models and closed source models is like shrinking down like so fast and we can't expect that super soon that this like privacy preserving um [clears throat] infrastructure to like enable the best like uh AI agents in the world you know >> in my opinion.
>> Yeah, I agree with PR actually. I think we are much closer than people think probably. Um but the key shift is that I think privacy has to become part of the routing decision itself right not an you know after sort.
The next step is adding you know trust as a a first class routing decision dimension alongside the existing dimensions like model cost, latency, model capability and so on. So I think this is where confidential computing and attestation become important. hardware backed you know trusted execution environment can protect the workload and data in use and uh independent attestation makes that protection you know verifiable and uh automatable I think that's the next step I think yeah >> yeah I I I tend to agree with the with uh what the other panelists are saying too but also I think that a big part of this is going to come down to what are the the use cases too um that that are finding the market fit uh and and because I think that the tech is closer than people suspect um to actually being able to do these things. And I mean like with what we're doing at Brave, everything is basically handled with privacy first and and user first kind of as our our principles. And so um we're able to go out there and stay competitive um with the with incumbents and new uh people coming into the market. So I think that um a lot of this too is is around finding the fit but I think that you know that's the harder nut to crack right now um compared to like technical challenges. I mean and we're also learning a lot too. I mean we we've given the ability for people to bring their own models and to do auto selection and and that and the amount of people that are choosing auto is it's um higher than what I think a lot of people would expect to.
>> Yeah. I I I think you know people are are I and I find myself some of this is personal anecdote you know like I I find myself choosing auto more often on some of the these models because well sometimes I just want a quick answer and and you know sometimes I wanted to really think through and then also like I am starting to get a little bit more token conscious whereas you know previously it was like let's just token max you know foot on the gas here let's let's burn all the tokens you know there there's different considerations and so I think you know who's going to be more intelligent about its own mo its own token consumption the model itself and and letting the model sort of choose okay you know what is the best you know the best model for this um is probably the best way to go about it and so you know when when we're thinking through this you know Jerry you mentioned private order or private model routing maybe you could just unpack a little bit more like what this means because at this point is it as simple as like I'm going to hit this general model, it's going to, you know, route my my query to, you know, whether it's the most advanced or the most thoughtful or the instant, you know, version of the model.
Is it that simple? And and where does the privacy get baked in? Maybe you could just expand upon what this private model routing means in context.
>> Yeah, sure. Um, from our perspective, from intest perspective, I think, you know, private model routing means the um routing layer does more than just pick a model, right? It also evaluates the trust posture of the environment where that model will run. So the router I think you know should consider questions like you know hey you know is this model running inside a confidential computing environment and is the workload or the model the non good workload. Yeah you right we have measurement of the workload itself. we we we we should ask hey you know is the model the known good workload uh has the runtime environment been tested and uh you know match the trust policy um so I think the important point here is that routing should not uh require build trust uh blind trust I uh in an you know uh infrastructure layer right or infra blinder trust in infrastructure provider it should be based on evidence which can be you know independently verified. So I this is uh you know our understanding. Yeah.
[snorts] >> Got it.
>> Something sorry a bit something I just want to say like on the other routing capabilities like this is still like an ongoing research discussion because it's not easy to say like for a given query uh is it do you need like a small model?
Do you need a big model? like it's very like arbitrary right it's like who decides that a question is like easy or not so I think there are like different personas right like usually like soft engineers most of my friends they would always use you know the best model like f like yeah I wouldn't say things like yeah the best models in the market right now with like the the highest effort you can do um and some would just like you know they just want the work to get done like if it's just like web search and stuff like this then yes it's fine to use these light models and just you know just pick auto because you're you don't really care if this request is not routed to the most you know competent model. Um I think it's just it's very different personas based on your different needs. But uh and I just want to say like auto classifiers like auto routing is still like TBD like you can see like the biggest um like the biggest providers in the market not all of them offer like auto because it's not it's not super reliable at the moment. Yeah.
>> Right. Very subjective. Yeah. and and I think it was Luke who mentioned you know there are uh very it's very case by case it's very case dependent upon you know what the user is trying to accomplish and so Luke given that you know Brave is you know in production and you guys have you know you know tons of browser activity you have a vantage point I'd imagine actually I'll frame this as a question do you have a vantage point into what people are using this for can you see you know the the and do some data sort of uh indexing on okay most of this is you know enterprise flow or most of this is user flow and you know they're they're interested in you know these sorts of questions or these sorts of queries or is it totally obuscated because it's you know brave is private at the browser level do you guys not have any insight into you know user data >> we spent a lot of time building out private uh uh privacy preserving analytics like uh especially like way back in 2017 18 even. Um and so you know having diagnostic data is super important and we know like aggregate things around like query volume types of queries things like that. Um we also know too like on the search side like how many queries are are coming up with an AI summary versus not. Um and then on the uh on the LEO side or browser AI assistant where uh a lot of the model selection takes place. Um I it's interesting because you know we have different cohorts of users right we have our main browser that's kind of the release that you get from the website but we also have a beta and a nightly channel and those nightly channels are are more the like the token maxer uh user cohort they want the latest they want you know um they're they're being very manual about things um and so we're seeing it really depends on the channel uh and and it depends on the use case but I I think that you know it's it's all over the place right now. I mean like um you know we see a lot around search but we but we also are trying to build in features that uh we can learn more about the types of use around as well if that makes sense. Um but but yeah it it's not like you obviously have like things like coding and stuff like that that people are using and our vantage point is interesting too because our search API is used across AI for a lot of things right like it's a default in openclaw and so we see a lot of um a lot of people like people that on the dev side that are working uh around our search API as well. So we do have an interesting data uh we an interesting kind of vantage point. I think a lot of what this industry is doing right now is catering to developers because that's where the innovation's happening and uh until you hit like a major use case that's going to get the like market fit level where they're going to lead. I mean you're starting to see that in some places but but it's it's still going to be catering to developers for a while.
>> Yeah. Guys, a a general question here is something that I've I've thought through is um the idea of locally run models because it feels like to truly get privacy preserving AI where you know our data which I think we already know just by you know things that come up when people really dig into these terms of service they realize that you know hey your data whether you like it or not you're going to get sold to advertisers right or your data is going to be available and you know if there's a uh you know something comes in from a law enforcement agency or something. This is subject to subpoenas and you know like these are real considerations. People don't want their data, you know, out there floating in the abyss. And it feels as if the only way to truly get privacy preserving AI is to eliminate the cloud environment entirely. If we're sending our data from our local machine up to the cloud, up in the cloud, very funky things can happen. How realistic is locally run AI? and and how important is it to run these models locally in order to preserve privacy?
>> Open question for for everyone here.
>> I mean, if you want to run the best models locally, uh this is going to cost you like a lot of dollars. Like it's I don't know the exact number, but it's like it's more than 100k. For example, if you want to run the full like JM 5.2 2 or Kim K3. This is like it's tens of thousands of dollars of like upfront cost you need to do on your hardware side to be able to run this this like these big models. And so this is where computing kind of lands, right? We offer the same guarantees as running models locally uh but in the cloud. So you as a user, as a a customer of this service, you have like a cryptographic proof uh that actually there is no way for neari or any like any of like privacy preserving solutions to actually see your prompts and the responses from these AI models. And this is where actually most of our clients are coming from because they realize that if they want if you want the best intelligence, you actually need to have like big GPUs which you cannot run at your place at least for now. I mean we've seen that consumer like consumer grade hardware is getting better and better but it's still not there for you to be able to run like Kim3 or yeah big models but if you are if you want to run like simple models like the fast ones then yeah it's possible to run locally but you you just lose a lot of intelligence that's the downside.
Yeah, Jerry, any anything to add there in terms of how you guys are approaching this from the Intel side? You know, given that maybe you've got, you know, a a focus on hardware, how are you helping to enable, you know, confidentiality? Is running things locally, is it realistic or is it even important in order to get privacy preserving AI?
Uh I I I think you know uh uh I actually agree with what what Pia just said right um from capability perspective it's it's probably doable you know run some model locally right but it's a confidential uh uh wise um it's still a challenge I think even you run some model in the local environment it's still the challenge you want to make sure the model you you get from whatever wherever you get. Right? You need to make sure it's the non good model. It's a a trusted model. Right? At this point, you still have the need to somehow to uh you know uh verify um the model uh you know is the the model you want to run, right?
It's there is no like back door this kind of thing. All those are uh about how we measure the running environment, right? So at this point confidential computing technologies still apply even in the local environment. you still want to run everything in a hardware backed SQL enclave and you know whole environment uh measured can be independ independently verified right I think it's still the same challenge no matter where you run you know model you know on prem locally on cloud on the edge the same thing it's the same challenge yeah >> and yeah >> sorry look I'm I'm quite curious so do you have like public data of you know the proportion of people because you mentioned at the beginning of this panel that you enable LEO users right to use their own models from LEO which is your AI assistant inside the browser. Do you know like the ratio of people going for this or going for like t solutions that that near for example has on the brave video as a small minority cohort like most likely single digits because most likely from around the folks that are in our nightly channel but I think that this topic too is one where we might find ourselves having a completely different answer in a couple of years um because of just how local models are evolving too. Um, you know, I have a a opportunity to talk to a lot of different specialized uh uh folks that are working on, you know, different types of hardware and things like that where they have to run local and it has to be very uh uh privacy conscious and um this is all early like on the local side I think and and right now it's just doesn't make sense to do it all. I mean everything that everybody else said earlier around costs and and and trade-offs is like very valid basically.
>> Okay. And so and what what is the biggest bottleneck here? You mentioned like this thing could change dramatically in the next couple of years. Is it just you know user education more people coming around that you know their data is being sold off and they they need to take control of that? Um, I think as more people start to use more of these LLMs and they transition from something just like raw search in a browser, they'll realize, you know, like there's people using it as their therapist, like they're really confiding, sharing very private information about their lives in these things. Is it just a matter of people coming around to the idea that this is being shared and they need to take that into their own hands or is there something some other bottleneck whether that's a technical bottleneck or or a non-technical one that you know we will ultimately solve going forward? What what are some of the limitations and the bottlenecks right now to turn on these features for more and more people?
>> I mean I think I think one and I just jump in quick on this because I've got kind of an interesting perspective on it. I think one like the people are becoming more educated on this stuff.
That's true. Um I we're starting to see that in other areas both with a lot of these like chat control types of things, air verifications and and even things like this whole deflocking thing like privacy stuff is becoming more of a thing. But I think that on the business side it's becoming much more relevant.
Um where people are starting to see that there are real liabilities in um in just kind of hoping everything works according to a policy. um especially when a lot of people start to observe how these policies are actually working uh in practice when they're evaluating the technology. So I think that those business risks are adding a lot more momentum towards um getting these solutions to market faster. Uh and there's a word that keeps getting replied here around like you know verifiability around and verification around um you know making making sure that that what you're using is is what it really is and and I think that's super valid too. Um I I I the the technical challenges are are obviously there as well. But I I think that whenever you have this forcing function of business concerns and business risks around these things, it helps to accelerate the adoption on the consumer side too. Um and the groundup adoption from the consumer side going to keep coming becoming a thing. I mean like when we started in 2016, people were questioning whether people even cared about privacy or not and and you know he's got like 120 million monthly users now and so a lot of people move with their feet for one reason or another. But when you really start to see this stuff kick in, it's when businesses have to be concerned because it really, you know, and they should be like I mean and and the point about the the the waiting room and and those things earlier or the doctor's office, right? Like all these different uh uh contexts in which people are using these things become liabilities across the board, right? And and so we even did some research around this around uh um prompt injections when uh agenic browsing first started to hit and people were just kind of letting things run wild and there's some serious considerations and and risk factors that people need to weigh in. Our our research team put out some uh uh research around those threats etc. So um I I think that the business part is is helping to accelerate this a lot. Um I I think that the innovation around cloud compute and uh and add astation and the TE work that that's going on is is helping too. Um just my own personal perspective on this.
Yeah, overall I do really agree with like look like at near AI like we see the demand for like privacy preserving LM solution like increasing a lot like u when we started doing this like two years ago the demand wasn't really there like we had to you know really pitch to people what is like what what are T's uh what is what is a secure enclave what near does everything and now people like come to us right they are like oh we need like a T solution for like our LLM needs basically They need AI inference that is privacy preserving. And we really see like the market starting to adopting this. And it's it's super cool to see because like finally people see that there are gaps in like in the biggest players at the moment. And it's just it's it's a bomb waiting to to explode. Like it's just they have so much data about everybody at the moment.
It's insane. Like everybody just shares as you said like Robbie and Luke like people share everything in there. like it's so so dangerous like if these things goes out like this data can destroy lives and stuff and I think people start to realize and that's u yeah it's just it's going to happen.
Yeah.
>> Yeah. The the last question I have for you guys here is you know uh now you know we understand running these things locally is just too high of a barrier to entry whether it be for cost or scale or setup. what is the low barrier, you know, to entry, easiest way to get set up so that I can continue using and and our community, our audience can continue using the tools that they're accustomed to using, but they can do so in a priv privacy preserving way. Is there an easy tool, add-on, hardware piece that they can now just incorporate into their workflows so that they can flip the switch and turn on privacy?
>> Gosh, am I going to show my own bags really hard here? I just you know use Brave. We have like Leo built in uh and our privacy uh model is like best in the market and uh you know you find and we also have an ear to the ground with our users too. So if there's stuff that they want we listen and and kind of implement that but um you know I think that that's me kind of selling my own bag here a little bit. But but yeah been here doing that for a long time.
Yeah, for sure you can use uh Brave Leo, but also I want to say that you can use near AI which uh we are one of the providers for Brave Leo and uh we empower you to do basically anything. So we have this API and you can leverage all these like open source models that we host inside and so you have the same guarantees than running uh everything locally and you just pay per token and it's uh much cheaper than using the the bigger the bigger players.
>> Yeah. And uh Jerry, what does your stack look like?
>> I think you know uh from Intel's perspective, our hardware technology like Intel just domain extension as known as Intel TDX technology really enable the like lift shift like deployment model uh for all the confidential computing workload include a confidential AI workload model this kind of thing. uh this significantly lower the technical bar for any team they want to you know uh deploy uh the you know um confidential AI workload right so they don't have to redesign their stack this kind of thing so basically our technology I think make make the privacy as part of the infrastructure provider layer itself right this definitely reduced I think the cost and the complexity to you know for people to uh to want to try that out. The other thing is on top of that you know uh besides the hardware technology you know our group is also uh uh delivering a product named the interrust authority which is uh you know independent uh testation service. the service itself is free, right? So uh so for uh we we work with uh the partner like NIA AI and a lot of you know major CSPs and also Neocloud providers to integrate that capability into their confidential computing uh service which allows the you know all basically uh everyone you know to try uh the confidential uh AI uh u private preserving computing. Yeah.
>> Yeah. [snorts] Fantastic guys. I think that's all the time that we have. Um really appreciate you all coming on. This is a you know a deep deep uh topic and it's also an extremely early one. So all of you guys are pioneers in your own right.
Appreciate you sharing the insights with us and and uh sharing your unique vantage points. Uh really appreciate it.
Hope to have you guys on again soon.
Talk a little bit more about the AI stack.
>> Thank you.
>> Thanks Robbie.
>> Thank you very much guys. Thanks so much.
>> Byebye.
All right, guys. We are going to keep on moving along. Uh having a fantastic show for you guys so far today. Let's check in on the chat. How are you guys doing?
Uh Alpha Rattle, the focus on privacy is the part of what I love about Near.
Yeah, I mean it's pretty incredible. I had really no idea that uh Near AI was collaborating uh with the likes of Intel on the hardware side, Brave on the browser side. Uh Near AI is a uh provider uh of confidential AI to Brave which is pretty awesome. Uh very cool to see that integration on the on their uh LEO product. uh near AI as a brave LEO backend with TE guarantees is quietly what the EU AI acts high-risk system provisions were written to incentivize.
It's pretty awesome. Uh pretty cool to see how Near AI is getting implemented into uh you know you've got uh LEO Brave and then you got Near AI on the back end. You got the TEES from Intel. Uh pretty pretty awesome to see. Lunaeth.
Luna is a crazy name. Anything Luna Teruna these days, data that can destroy lives and people are still just handling handing it over for free like nothing happened. No, it's pretty insane. Uh people are handing all their data over to AIS. It's it's pretty nuts. That is why uh you know the privacy preserving elements are becoming increasingly important.
Uh so very very cool panel with Intel uh and Brave as well as with Near AI uh to learn about how all of them are working together. Uh let's see Akan in the chat says one of the things I love about Near is the focus on privacy. Yeah, very very cool. Uh as we said Near AI uh is serving as one of the backends for that Brave Leo system. Uh, and then you got the Intel TEES, uh, serving as that hardware enclave that helps support the system as well. Last thing we got in the chat here, cryptographic proof that nobody can read your prompts is a generally wild value prop and it's being explained right now like it's obvious.
This is the actual unlock for enterprise AI adoption and people are still sleeping on it. Yeah, I mean we heard how high of a barrier of entry it is to run these models locally, but um you know there are enterprises that are very very conscious of their data. Uh it's very important to them that they don't leak these confidential secrets and so uh these are the guys that have the resources and uh and the willpower to deploy and run these things locally.
That's where I think it starts. And then over time, we're going to bring these costs down. And I could understand how consumers uh get access to locally run these things as well. Over time, in the meantime, we got TEES and confidential browsers. Guys, we've got our next panel. I think this is one of our last panels of the day. Second to last panel of the day. Guard rails for Aentic Commerce. I see these guys are ready in the back end. So, we're going to bring them up in just a second and talk about guardrails for Aentic Commerce. You guys heard a lot about this concept here today. Description on this one is static security tests were built for a world where attackers didn't use AI. They do now. Yeah, it's been honestly it's just been sort of [clears throat] concerning, annoying. Not sure exactly what the adjective is, but it feels like the malicious actors have been using AI more than the ones who are fighting for good. Uh we're going to talk about it.
We're going to get guard rails for Agentic Commerce going here on the show.
So, let's go ahead and bring up our panelists and get this one cracking.
It's a full house, guys. You're on stage now. Welcome to the show. It's a pleasure to have you guys.
Um, and so we're going to get we're going to get the conversation flowing.
We're going to get we're going to have some fun here. I'm just going to ask some open-ended questions, guys. So, just feel free to just jump in, uh, you know, talk over each other, have a good time. Um, but as I mentioned, you know, we've got, uh, uh, guard rails for Aentic Commerce. We've got Ari from Fail Save, Zaki from Poles. I know Zaki's got a a million things that he's got going at any one time. Cameron from Near AI, and Ali from Ledger. Um, so we're going to we're going to talk about it, guys.
Um, you know, I'm sure you guys have seen uh all the it felt like there was a just crazy run of exploits. Uh, and it was right around the same time that some of these models, these Frontier models were coming out. Um, it feels like we should be using these models for good, not evil, not to be using them for exploits. Um, first, do you think we're now in a place where, you know, we can start to use these models and use them for, you know, agentic commerce? Like, I guess maybe the first question to ask here is what does the state of agentic commerce look like today? Because we've had a lot of theoretical conversations, but let's just get a little bit more practical, open question. What is the state of Agentic Commerce? I know it's general, but we'll we'll hone in and where, you know, feel free to jump in.
Where do you guys think where we are today? And where do you guys each fit in to the Aented Commerce stack?
>> Go for it.
>> Yeah. Yeah. No, I think uh right now, by the way, I'm Cameron from Near AI. I've been building with Near for the last six years or so. And uh where Agent Commerce is today is for digitally native things.
So things like APIs, uh I guess like agents paying other agents to do certain crypto-reated tasks even. Uh but I would say it's it's really started with digitally native um because you know all the primitives are there for it. Um but right now we are and also like we're starting to see uh examples with like Mercury and Brex like kind of hit like the mainstream uh with uh things like like smart routing uh right now with Abound is a company that we're working very closely with where we're doing smart remittance. So having an agent essentially call an API to get a foreign exchange price to do remittance at the best rate that they can get. So someone can talk in natural language and essentially make you know get a better rate for sending money abroad. And so right now it's still these like definitely financial sort of uh native experiences that are inherently digital but we haven't seen a ton of real world stuff. uh you know obviously there's some example like you know uh almost case studies with convenience stores being completely run by AI like navigating supply chain but honestly it's a little bit of a pet projects rather than anything like you know enterprise ready uh but we're getting there >> yep um Ari uh we were on stage I think it was in Miami and you were uh you know discussing the attack bench benchmark that you had that a lot of these, you know, uh, these models were starting to get raided against. Uh, we were up there with George. I'm just curious, you know, as we move this agentic economy forward, you know, and and again, like start to establish some guard rails, but, you know, again, you know, just try to limit the amount of exploits and use use these things for the power of good here. What have you found? You know, just give us a catch up. I think that that conference was a couple of months ago at this point. Have you seen some adoption and have you seen some teams start to put into practice you know some of these some of these models uh for the force of ultimately using these things for uh you know powerful technology uh you know what what people ultimately want here not exploits but you know agentic commerce moving forward.
>> Yeah. So I think uh a lot of the stuff that we collaborated on together with with Near uh was really around evaluation of of claws right different claws uh whether it's ironclaw open claw Hermes and um uh and while I think originally uh there was like a ton of of security issues uh with openclaw and and confused deputy uh problem uh I think that all of these plat platforms are are tightening down and it's really interesting to see uh frontier labs that are basically I mean uh if you look at uh uh open AI work there's claude work these are essentially claws just by another name so putting uh I think putting constraints putting permissions um around the actions that can that that an agent can take is is becoming more uh more mainstream It's Yeah.
>> Yeah. Uh Zaki and and uh uh Ali, you know, what what are you guys seeing on the ground? What what is the state? Do you think we're still in the you know, using there's, you know, AI in the hands of bad actors and that those guys are still uh using it uh more so than the guys that are, you know, the good guys, if you will, or generally >> I don't think that's true. Yeah, I I I don't think that's true at all. Um I think we're in a weird just in you know having having been I guess my my my like multiple takes are one is you are seeing a massive acceleration in terms of what can be done in terms of uh building onchain applications that are like hardened in a way that would never have been possible before. Um, you know, I would point to like what Zcash is doing with Ironwood, um, as a great example as probably I would say the kind of the the high watermark right now of what is happening in the crypto industry in terms of of like using AI to drive like defensive programming and building a a blockchain release that is very built around building things defensively, using AI to build defense. Um, but I think it's I think you have this I think that you're going to have a a whole generation of new DeFi projects that that also lever it. But I think that like one of the things that is also probably the biggest challenge right now is I think since October from like all of the experiments that I've done under a variety of different you know hats whether it's similier or um or the stuff I work on with Near um it's been very clear that um agents are very well positioned for you to access to like act for you on chain um like they can do they can do trades for you. They can enter into D5 positions for you. They can monitor those D5 positions for you. But the hard part of putting these things into production has always been security. Um like the the work that you know uh Ari has done on attackbench like the TLDDR of what it says is that everything we know like we do not know how to make an agent that reliably acts in the user's interest and isn't subverted by the internet. Um so then what the question is is like what are the compartments that you build into the system? How do you compartmentalize the system so that there is a defense and depth from between what the agent do like what the agent is sort of communicating and you know giving you a a you know telling you to signing on your behalf various use cases um in in various flows are there but how do you have defense and depth so that you aren't you don't get that subverted message that just you know sends all your money to North Korea How do we do that? Like uh I you know like the is it in the harness? Is it in you know the application providing some of these parameters? Um it you know I olly I want to hear from you on the hardware side like is there hardware constraints that we can put on these things to effectively harness them a little bit more you know securely. What what are you guys seeing on the ground?
>> That's the most interesting part. Uh I want to touch on what Ari mentioned because when cloud co-work and now open AI work came out. It's not being AI in the hand of bad actors. It's candy in the hands of babies. All of these new people building AI agents with all of these tools that the barrier of entry is so low that everybody is just shipping and the last thing they think about is the security. Everybody's talking about the infra. Nobody's talking about who's holding the keys.
I'm I'm super bullish on AI thinking on AI proposing but the fact that is missing is we should keep the humans in control as not as a policy as an enforcement. So the only thing that we always like ledger is a beacon of safety in the crypto world. So now we are pivoting hard to the AI side because we are seeing the same thing that was happening in 2021.
When you open Twitter in 2021, everybody was getting hacked. It was so like it was every day it was on the timeline people getting hacked. We're going to see the same thing unfortunately with the AI agents if people they are not going to take security more serious and that's where the hardware comes in because AI AI can hack software. AI cannot touch hardware and that's the part that we enforce very hard.
>> Yeah. Um, Ari, and I'm sure you've thought deeply about this because, you know, I I think it's probably true to your nature not just to put out the benchmark and then hands off like, you know, you put out the benchmark, you draw a conclusion, and then the idea is like, let's do something about it, right? So, if the TLDDR is like, hey, like there just isn't a harness, there isn't guard rails out there that can prevent things from the internet. What is the best that we can get? What would you suggest to teams that are looking at building out more of these agentic transaction flows? what can they do to help you know secure that system as best pos as possible?
>> A great question. Uh I think it's first starts with uh an objective evaluation of of your system. So um you know the you know this this is kind of the the motivation that we had together with near like when we first looked at evaluating different claws. uh most of the stuff pretty much all the stuff that we saw were just a bunch of static tests and we felt that that's that's really uh uh insufficient that the the in practice attackers have been already for for some time leveraging LLMs and doing intelligent attacks and actually seeing the inference that's being done by by the other side. So I think it all it first should start with an objective ailation of your own system. Um so attack bench is kind of is meant to do that but some either that or some variant of that where you have an adversarial attacker right so the the attacker is an LLM and it gets a bunch of turns and it actually has visibility into your agent into the into the reasoning of of your agent and then you know it'll it'll change tax it'll try different things um uh you know trying different configurations of that uh and then and there's no single s silver bullet so it starts So that um and then um defense and depth there's any number of mechanisms uh there there's some really interesting um uh what I thought was quite novel and clean approach that ironclaw had with permissioning it was uh with right permissioning had uh very strong controls on that but basically it's additively multiple layers defense and depth starting with the v uh like a clean visibility of your own system understanding it's uh you know don't just run a bunch of static tests, they passed with a particular LM configured, you call it good and ship, right? Um, so it's just it's it's uh it's a combination of of of these different things. Yeah.
>> Yeah. Um, Zucky and and Cameron, what what is the status of Ironclaw? Because uh it felt like it came out and, you know, OpenClaw was kind of taking the world by storm. Ironclaw provided this, you know, more secure harness for Claude.
what has been the what has been the movement on it uh in the last few months?
>> Um I could I could I could take my perspective on it. Um, so like one of the things that ironclaw security model sort of enables that is I think kind of unique is we're you're we're is it allows you to run these kind of organizational level clause rather than just the uh the sort of individual user claw. Um and so you know like one of the things that we've always kind of you know one of the things that's always a struggle with software is like security is not a feature. Uh sometime it's it's table stakes it's trust but nobody ever adopts a piece like there's a very small group of people who are like I will take the worst thing for the more secure thing. They are like I want the best thing but I want it also to be secure.
Um, so what you got to ask yourself is, okay, we we invested all of this enormous amount of time and energy in building all these security features into uh ironclaw. Like where does how do you take those security features and then do a product that is 10x better than um what else is there? Um and so the product team um yeah, I've been a I'm a maintainer on Ironclaw, but there are lots of other people who are doing far more work than I am on this right now. Um uh the product team has been really trying to push this architecture of the sort of organizational level instance and like you see you know the the um the both in the open source world like paradigm has this centaur thing and then anthropic has this claude tag um kind of concept. We're trying to build out this like organizational level ironclaw experience. So basically it can be you can deploy it to like an entire company um and it can operate in like a coherent manner. Um, so that's that's like kind of the where where we're going on the product direction.
>> Yeah, exactly. Like the the multi-tenant instance is what we call it internally and you can spin up this multi-tenant instance with specific skills, tools, workflows per department. And so if I'm talking to a government and they, you know, are responsible for pension payouts, like everyone who is automatically working in pension payouts already comes with the exact tools that they use and workflows to to make their job easier. And we're currently adopting this internally inside of Near Foundation. You know, the goal is not to replace the team. Let me be super clear about that. Uh the goal is to empower them to do a whole lot more work than they were able to do before and kind of take the benotney out of like you know filling out the legal request form to you know start a contracting process and have most of that be agentic depending on the call transcript that you had with a partner to then even do like forecast calculations to determine like the deal value to determine how this sits in a pipeline. like these sort of things take you know teams a lot of time and energy and this can all be automated and it can all automatically be stood up within an organization's multi-tenant instance. So where things are today is we're making it a whole lot more reliable. We're introducing like self-learning loops so it's constantly learning on the traces that it's developed and then even giving users the option to I mean I don't know if we should talk too much about this but there's no open source repo um so I guess we could is uh you know allow people to use those traces to improve their workflows and even potentially like monetize them later. Um, and so there's a lot of uh really interesting stuff being baked into a new release uh that is slowly but surely coming together. And then we're just taking this and applying it like we are showing not telling um through use cases that require this level of sensitivity because every every large organization wants these claws. It's just that there every time it goes to the CISO, they're like absolutely not. We can't touch this with a 10-ft pole. It's going to leak our credentials to the LLM. it's there's all this hallucination risk, you know, all this stuff. So, we're just proving it out with partners. Um, one of them again is this like abound use case which I think is super compelling because this is an agent that is actually using money to send it abroad to non-crypto native people just like everyday like people that need remittances and then we could just rinse and repeat this across like every single remittance company or something like that. And so um there's a lot of exciting stuff here and I would rather show not tell and that showing is coming very soon. and and so it's the idea here that um by and I want to just make sure that I understand the the the implementation process here but when you say trace you know this is essentially the concept if I understand correctly that you know you you put in a prompt and you sort of know what you're going to get out you know like right now LLMs are almost entirely probabilistic you know I could put something in you know Cameron you could put something in and we could put in the exact same thing but the same inputs could lead to two different outputs because it's a probability that we get you know a certain thing each but by using this trace is that taking these LLMs and making them deterministic for these large organizations so that there's a little bit more of an understanding in terms of you know what they can expect and that therefore you um you take out some of the variability of these models and therefore that that determinism creates the guard rails I'm happy to talk about this top. You go ahead. Yeah.
>> Yeah. I I I would position it very differently. So if you if you look at like so there's this whole concept of post training and post- training like sort of took off um in after at the end of 2024 um with uh OpenAI released 01 and uh uh Deepseek V3 um came out and or Deepseek's R1 came out um which both sort of showed the whole industry what is possible with essentially giving when you are like sort opt using developing the reasoning capability and the ability to be a a useful worker uh based off of lots of examples of work. So right now this is there's the the way in which all of the labs that are out there are all improving uh the mo improving models for different work tasks and is uh is by feeding them examples of work. Um and there's this whole data economy that has emerged about it. And then there's this whole also this whole world that has emerged um where you have teams like thinking machines and uh prime intellect uh which have built entire platforms for companies to take open source models and then their own proprietary traces and build models that are very good at their workflow. Um, but this basically right now is where we're at in terms of uh how how we're this is this seems to be the sort of industry consensus of how we're getting to AGI right now. Um, is sort of uh somewhat brute forcing our way through lots of examples of doing interesting work um and then having the models post- train over them to get them to the point where they are become sort of human equivalent workers. Um so this is this is this is like really fundamental just sort of just the whole path to AGI that we're on.
>> Okay. Okay. So the the ter the trace the deterministic part of this is actually uh separate from the model and then we apply it. you imagine like you like are like okay so like I am trying to take um you know model N like version N and people are using it uh or people and people are using frontier models you know proprietary models all these things it's all the trace is the prompt the response from the tool calls that like the model uh selected and then the question and then the so like now you're trying so and then any corrections that the user made. Um, and and corrections are super valuable because when you're training the next version of the model, what you're actually asking the question is, >> okay, like the model made a mistake or like got to the right maybe got to the right answer but not in the most efficient way or made a mistake and the user had to steer them back on course.
Maybe they had to steer them a lot. Uh, maybe it never got there. Um, and now if you can take that trace as a target for post-training the model, you can make the next version of the model be like, okay, did it do better on this prompt or did it do worse? Um, and then, you know, adjust the weights until it does better.
And in theory, we could actually use attack bench as like kind of this, you know, we could like send the adversary, you know, we could judge it and then we could use some sort of uh like fixes that we implement and then we could trace that again. We could in theory get more and more secure models because they're better at defending against these adversaries.
>> I want to hear Ari's take.
>> Yeah, I mean it's attackbench is actually it's a combo. It tests both the the the agent and the LLM. Uh it was really interesting the the findings we had. It was uh kind of unexpected. It's it was really the the uh you know in in some scenarios you would have for some type of attacks it was the LM that came through and saved the day. In other cases it was the agent. In other cases both missed it. Um so I think it can you know for for what you described certainly it can uh uh it can uh help in uh like you know after after you've post-trained the model and it's in the context of this particular agent um you can you can run the same workflow uh adversarial workflow and see what happens. Uh it can be used as a tool that way. Um but uh but you probably also separately, you know, uh you probably want to have a a uh uh a separate test bed that's just focused on the LLM as well, right?
>> Yeah. H I mean this is this is deep deep stuff. So I I I feel like we could we could chat for uh you know significantly longer time about this. I the last question I have is just around bottlenecks like uh you know all of all you guys are here you're building through you know tough market conditions out there um what what is the limit limiting factor right now you know uh just open question to everyone you know where should we where should we be focusing our attention to ultimately you know make these things more secure enable the agentic economy on chain and and increase user adoption 10fold is Is there a common bottleneck that we can be you know focusing on or generally what what what do you guys think is the limiting factor?
>> I would love to jump in here if everyone's okay. One of the things that we are seeing uh just to uh continue what Cameron was saying is showing rather than telling. If we want the agent commerce to widespread scale in the scale that we are expecting it to be, the security should scale at the same speed. But we are not seeing that at the moment because every time security is the boring subject and everybody goes to security when they have a when they encounter a hack encounter an exploit and the compromise part is the hard part. So building is building in public can help that bottleneck a lot. Every one of us every other brand we are focusing on security.
We are building very nice tech. If we can share it more, show it more and make it more public, the eyes of the normal people who are playing around with agents, they get more aware of what's happening and what's coming and keep them more secure and go out go out and look for the infos that are being built and they don't know about it. I think that's a major bottleneck in the security.
>> Yeah, absolutely. And guys, feel free to jump in. you know what what do you guys think bottlenecks limiting factors?
>> Uh there are a lot of bottlenecks. Uh there is the whole identity reputation piece that I think is only really going to be important once we have useful agents. Uh that you know and that useful agent is requiring a sort of new UI because candidly it's a lot easier for me to go buy something on Amazon than it is for me to go chat with my agent to make sure it's like buying the right thing. And so we completely need to redefine like the context window and like how memory is managed to actually make that agentic experience personalized because if I have to do more work for this agent to act on my behalf, I'm not going to do it. Um so like I'd say memory is like one of the biggest issues here. Um and it is being addressed through these agent frameworks. And then another huge issue especially for crypto and this I've not seen an elegant solution here is around chargebacks because humans hallucinate.
you know, I buy the wrong stuff all the time. Uh, let me make if I was paying in crypto, uh, and the AI model is going to hallucinate as well, but if I was paying crypto, that money is gone. And so, I need some sort of chargeback mechanism or some level of insurance or the vendor needs to be able to um, also accept payments privately. No legitimate like vendor wants to have all of their customer user information like on a public immutable ledger and how much things cost. And so I think this is where like confidential intents really comes into play. Like and vendors should be able to accept any asset on any chain. Like if I wanted to pay New York Times and New York Times only accepts USDC on Salana and I only have Dogecoin on Dogecoin, I should be able to my agent should be able to pay for that article or pay per crawl or per scrape via my Dogecoin. And that's where Near Intense and kind of how this whole thing kind of plugs in together. And it needs to be private. And so you start adding all these things together with solid benchmarks and eval frameworks. It starts with the benchmark which is what I love the fail safe work. Um but that uh actual agent framework is critical to make sure like secrets don't leave the uh the like doesn't touch the LLM. Uh make sure that like there's all this stuff. Um but I would say like the main issue is probably charge package insurance uh because these things are going to mess up and transaction simulation. Saki, I don't know if you want to plug in there, but um it's a big >> Well, yeah. I think like one of the, you know, I think the ledger people are obviously very knowledgeable about the challenge of that blind signing has been as an industry for all of these years.
Um, in some ways, AI agents kind of compound the problem. Um, because you have this kind of what you see is what you sign thing. Um, and you can imagine attackers that are going to try to subvert your ledger is telling you, you know, you're sending money to your Coinbase account and like it's actually to another address. Um, so there's a lot of new attack surface area and there's a huge need um to sort of have to build a secure channel to the user to say like, okay, before I say I'm going to sign this, like where did what is this doing?
How do we know that the address is involved, the contracts involved, the effects of the transaction are um and things like near intents actually like make this much easier in many ways because intents have like very encapsulated effects like you're like swap you know this much uh USDC for Zcash or Bitcoin like it has a very confined possible set of state transitions whereas a transaction that interacts with the smart contract actually there it's a very challenging thing um so like one of the parts that you need to have is like a is an ability to sort of know your contract, simulate the outcomes, all of these things in a way that then gets sent where the user gets clear visibility. Um, and you can do things like, you know, make sure that the context window that is presenting the transaction to the user doesn't did not share anything with the context window that generated the transaction stuff like that.
>> Yeah. And if I could if I could add one u I mean there there's uh like Cameron said there there's a lot of different uh u challenges but if I had to pick one u the thing that kind of um bugs me the most is the uh uh so the current mechanism like how do you deal with uh tough decisions? Well you ask the human to answer it for you right? So give me permissions to do this and this gets abused to no end. And then we've seen you know some variant some application of this for the last like you know three decades it was just in different context as the human eventually the human just just clicks away and and all is good but it's is especially uh important in in uh in this context because uh you know you have an intelligent agent that's making a decision and you don't want to be I mean if you ask the human every time then you're degrading the the whole value prop of the agent itself. So I think there's there's a lot of interesting uh possibilities and and and work that could be done in that area making you know automating that decision uh having a uh an agent uh if you will that that is that is specifically trained to make the right kind of decisions for you.
Easier said than done. Uh but but I think that's that's a particularly key key area to address.
>> Yeah. or or at least an agent that is able to say like, "Hey, is this a decision for an agent or a decision for a human and and directing it uh accordingly." Uh guys, thanks so much.
Uh really appreciate you joining today.
It's been a pleasure and um yeah, I mean, this is something that I think is still extremely early and we're going to hear a lot more about. So, appreciate you sharing your insights and uh we'll see you all again soon.
>> Thank you guys.
>> Thanks everyone.
>> See you.
>> Cheers, guys.
All right, we are on to our last but not least, our final panel. Um, pretty fascinating conversations there. I mean, it's tough to really understand, you know, at some points it's like, hey, you know, we're ready to go and transact all these things on chain.
Other times it's like these things don't have any guard rails. like we need to do some serious serious um serious uh work in order to get these things up to a place where they are secure enough to make transactions on chain. Um there's uh there's some real considerations here.
We're getting some gene in the chat. Uh let's see. Liam Foster adversarial red teaming as a continuous post training loop is actually the most interesting security primitive I've heard in a while. long volatility on any infant play that figures out automated attack defense cycling before the big labs do.
Yeah, it does feel like there's some real innovative work being done here and the big labs could be uh you know taking it uh and paying for it. Uh Alex Webb 3 says brute forcing our way to AGI by showing models enough examples of human work. That's the plan. We have named this a consensus. Uh yeah, I mean you guys have seen all the data labeling everything that happens there. Uh Johnny Pan in the chat feeding models examples of work to make them better workers is just apprenticeship. We invented this [clears throat] in 1400. Well um you know AGI or AI got invented a couple of years ago at year zero and uh you know now we're we may very well be in year400 of AI models. Uh Alex companywide AI agent deployments with coherent behavior is a nice goal. coherent across whose threat model exactly.
That was one of the cool things about the benchmark from fail safe. You know these guys are doing uh the benchmarking to understand you know what is coherent, what is the threat model, how much of a threat really is it. Uh as far as you know what we do after the fact uh to protect against those threats, you know that that's a whole another conversation. Uh, you red teams in the chat. Defense in depth is correct and also every team ships after one config passes. Um, lots of good comments here in the chat. Hardware constraints as the enforcement layer for AI safety is genuinely interesting. Um, very very cool to see. Zcash Ironwood is cool. Uh, lots of great comments in the chat.
Appreciate you guys rocking with us. Uh, we are on to our final panel of the day.
Last but certainly not least, um lots of really really great improvements happening on the AI agentic front. Uh we're protecting these things. We're making them more secure. Near AI is leading the way. We also have Near Intense leading the way. Uh we've talked about, you know, the intent world, uh how Near Intense is is working on this.
We've also talked about the harnesses.
Um we've got uh George and Gordon in the back. We're going to bring them up in just a moment. Uh if you guys were in Miami, you saw George and Ari were on the stage. We were talking about Attackbench. We just had the conversation on guard rails. This next panel is going to be around private payments and uh we're going to get into it uh and we're going to get into specifically what an agent pays in whatever it's holding. A vendor gets paid in whatever they actually want on whatever chain they prefer. Near intents and circle are building the infrastructure that makes that exchange seamless and private. So let's go ahead and bring them up. Uh we're going to get into our final panel of the day. Uh because we've really had um a conversation all across the stack. We've talked about, you know, what it takes to make these authentic payments private.
We've talked about confidential intents.
And so it feels like this is a nice wrap up to talk about what's practically happening now. We've got George and Gordon on the show. We've got them up on the stage. Guys, it's uh it's a pleasure to host you guys. Welcome to the show.
>> Pleasure to be here, Robbie.
>> Thank you for having us.
>> Absolutely. Thanks for uh thanks for vibing out with us today. Uh Circle, you know, obviously Goliath in the payments and the stable coin space. George, you know, you're you're one of the leaders on Near AI, uh representing both Near AI and Near Intense. Um and so, you know, this feels like a good wrap-up. I actually want to share a thesis with you guys that I've been I've been thinking about on the show, and I'm curious how you guys see this, and you know, maybe this is going to lead into our stable coin conversation here. Um, the idea that I've been kind of toying around with and I'd love just a critique from you guys is that, you know, the consumer behavior on the stable coin front. When I go buy a coffee, I tap my card, right?
Everyone's used to tapping their card.
That consumer behavior is probably going to stay the same for for a long time.
I'm just going to continue to tap my card. On the merchant side, you know, they get shown their bank account balance. When they make a sale, their bank account balance ticks up. They don't really know, okay, my bank is in, you know, five-year treasuries or 10-year treasuries and they have mortgages out to these payments and whatnot. All the merchant sees is this number on the screen when they log into their online banking and all the consumer does is tap their card. That's their consumer behavior. It's on the back end. It's up to the banks that are going to decide to manipulate their balance sheet in such a way where they're going to start to allocate more of their assets into stable coins. not going to touch the consumer behavior, not going to touch the merchant behavior. All of that is going to be obuscated away. But we can still onboard trillions and trillions of dollars of stable coins at the bank level. And to to the consumer and to the merchant, all of this is behind the scenes. We just see number that is what we know as money. It's just the number on the screen. The stable coins are handed by the people that work on the balance sheet. Do you guys think this is, you know, a fair a fairs? Do you think this is going to play out on the stable coin side? Maybe we can just set the tone here for how much of the adoption curve is going to be taken by the institutions versus versus by the consumers or the merchants when it comes to the the stable coin adoption. And we'll let that set the tone for private stable coins, private payments, how this these things play out over time. What do you guys think?
So I'll start by you know coming from uh a regulated stable coin side that the most important part as a form of money is uh uh the acceptance of that instrument when no questions asked.
That's kind of what defines uh to be truly a payment instrument that the merchant receiving it does not question is this instrument going to be safe? Is it going to be stable? Uh so that's why you know circle operates well uh within the regulatory parameter and within the uh broader financial system to offer that onetoone redemption to offer that seamless experience across uh also multiples of uh jurisdictions across the world multiple financial centers and that is the seamless experience that we hope uh certainly eventually consumers will feel as well when they pay and the merchants will feel as well when they see their bank account [clears throat] or stable coin account or neo bank account take up hopefully take up a little bit more because uh uh it will make things more efficient and uh it will be overall cheaper for merchants uh so you know we're providing that core foundation in terms of uh safety and payments >> to what Gordon's saying I agree with your premise Robbie that I think one of the the statements I like to meditate on some sometimes is Marshia McLuhan's famous saying that the medium is the message which is like this idea that essentially the interface with which we interact with products shapes us and in turn we shape the tools that we end up using right so your example of like uh buying coffee by tapping a phone um to actually purchase the coffee I think that is the minimum requirement needed for really good payment products to take off it's got to be at least as good as the current state of the products I think there's there's an opportunity to make things substantially better with both stable coins and private stable coins both in terms of the user interface as well as what happens behind the scenes for merchants as well as payment processors. Um but that's a long conversation we can have at this point in time or or at a later point in time as well.
>> Yeah, there there's you know very real limitations to the current dynamic in which you know that tapping card takes place. You know, I still go in places, usually smaller shops, and they say, you know, it's a $10 minimum to use a credit card or, you know, if you know, you use cash, you get a discount. If you use a card, there's going to be, you know, service charge or these sorts of things.
And it feels as if um you know, bringing down credit card processing fees, uh is the way to do that. And and how do we bring those fees down? Well, it's it's stable coins. And so um you know we could we could make this a regular you know stable coin conversation but I think there's plenty of interesting things to talk about on the agentic side as well. Um I guess you know the the the one limitation is that the agents themselves right we the previous conversation I guess you know just helped me realize like how you know we're still lacking guard rails and George I I I remember this from you know when you were up there on the stage with Ari and we were talking about these benchmarks and how these things you know are still susceptible to some of these adversarial attacks and how you know it is still somewhat insecure that these things are agentically doing this h what maybe you could just George you can just give us a a a quick status update where are we in the agentic world because I'm getting some mixed reviews on you know how much Agentic transactions have really taken off where we are today and ultimately what we can do to push things forward but first where do you see us right now what is the state of agent the agentic economy and and how safe how secure how real is it today >> I think the question about security and how real the agent economy is uh has to go back to kind of We're in a technology war both in terms of people building products and attackers of those products. And to look at it as a static state as a fixed point in time ignores the fact that what matters is the larger longer uh trajectory we're looking at in terms of actually how we are looking at building safe and secure products. One of the things that near has been sort of looking into and working on for a period of time has been formal verification.
formal verification of code, right? That is something I think we can make a decent amount of headways into actually making sure that products and software are secure. Another example is um working with folks like the uh the partners we've worked on in terms of creating really really top quality security benchmarks. It's really hard to gauge how secure a product is unless you have benchmarks for it, right? That's been something that we've worked on before and we're continuing to work on going forward to improve security not only at uh at the harness level but also evaluating what we can do in terms of the interactions with with the different models as well. To your original question about how much agenda transactions have taken off I think we're still very early >> Robbie right now it's at the point where we're building products really finding out where there's strong product market fit for those products. That's a question we have to answer in a definitive and strong way before you're going to see meaningful large adoption or increases in transaction volumes for for agents.
>> Yep. Makes sense. And these things are both extremely early on the stable coin front on the AI the agentic front. Uh but ultimately these things are converging. We see the trajectory is inevitable. Now it's just about accelerating uh this because we realize that this is technologically uh more efficient. It's innovative uh and you know as a believer in technology myself uh capital wants to be free. It it tends to get uh you know proliferated towards the most uh frictionless the most efficient route possible. And so I think this is this is ultimately where we're headed. We can all see where we're headed. It's just about how quickly we can get there. Gordon, from the circle side, do you agree? You know, is it still extremely early? What has been the traction that you're seeing? you know, you guys are big proponents X42, you know, you know, nano payments, these sorts of things. Where do you see us on the the Agentic payments pathway and and what can we do to accelerate um our our journey along this path?
>> Yeah, that's a great question. I I think we are early but I would caution that it is oftenimes really really hard to estimate especially these exponential technologies especially in terms of AI right uh you know I I think a lot of people especially in San Francisco believe that we're on the verge of AGI in the next couple years next call it one to three years type of range then afterwards anything could happen you would imagine that agents would be driving the a large portion if not the vast proportion of uh economic activities on the internet. When that happens, when you have economic activities and exchange of value has to occur at the same time because you can't just provide the services without collecting the payment, collecting the fees. So right now we are building the primitives you know as you mentioned X42 is something that we're really proud of because I think it's something like 99.9% of this 99 some% of X42 settlement happens in USC we've been actively working on agent wallet giving wallets to AI agents so that you can use it securely for payments uh nano payments another area in which you know paying for small value transactions whether it's for API calls, inference data, we just see that being a necessity for AI agents to function well uh in both the current system but also in in this future AGI world and I think it's just hard for human to grasp how fast things are changing this exponential curve that's about to happen. We're at the inflection point at this point.
>> Yeah. And I just looked up a quick uh x42 dashboard, x42can.com.
Uh I think this says 18 million transactions. Um it still says $860,000 of volume. Not a ton of volume, but a lot of transactions flowing through here. And I does I do think it shows, you know, the power here is in a lot of nano payments. There's a lot of there's a lot of opportunity here to to scale.
Um, and I think there are, you know, use cases that become available that previously weren't available. Uh, today we had people talking about, you know, streaming, uh, for API feeds, making payments based on calls. You know, this the same could be true if, you know, rather than the the typical, you know, monthly subscription model. We can seriously invert, you know, some of these business models on their head, uh, and and everything can sort of become ondemand. I think there's a lot of opportunities with streaming payments this way as well. um you know this becomes available with stable coins. Um and so uh as you guys are are you know continuing to accelerate in this direction it seems as if you know privacy is going to be an extremely important piece of this. Uh it's going to be essential for Aentic commerce. Um you know the these things are just table stakes at this point. Uh the other thing that I like to I like to think about here uh because I think it is extremely early but I I can't help but think that this is going to become more adopted over time is uh it is the product o over at Near that has you know it allows you to ask or essentially buy like aentic labor and you can use you know payments for these things. Um just you know thinking through some of the use cases here. Why is privacy so important and how have you been able to implement privacy uh into the stack so that you know as you're making these transactions private payments become ubiquitous and table stakes in Agent Commerce. Open question to both of you guys. Yeah, for on the near side, I think one of the things we see is that for anything that you do that is potentially sensitive, whether they're financial transactions or buying services, um privacy ends up being a critical primitive. Do you really want to broadcast your trades or the the services that you're purchasing or the goods you're purchasing to the broader world? I think none of us probably want to do that, right? So, we see that having that as a really critical part of this. The way that we've seen the critical elements of how do you create a better AI agentic economy is having a number of core primitives including privacy being a critical one. So one is privacy. Another one that we've ended up uh building as a core primitive is private inference confidential inference running through trust execution environments. We think that not only does the inference matter a lot but also the agent harness you use matter a lot. So that is the reasoning behind building R andclop Robbie which I know we've discussed before. So if you take a step back and you have uh transaction infrastructure that's capable of being private, you have inference that's capable of being private and secure and you have agents that are capable of being private and secure. We think those three components are the key three components of building a powerful valuable private and secure agent economy.
>> Yeah. Um, a absolutely and it's incredible to see. Uh, Gordon, anything to add there?
>> Yeah, I mean, we also very much believe privacy is a core primitive that's needed. Uh, especially because we work with so many of the regulated financial institutions.
>> Uh, you know, both people were exploring the AI side, but just real transactions, you know, yes, they were part of this larger DTCC test uh of set of tokenized asset transactions and USC was used as the cash. Of course, in any regulated financial transactions, you want to maintain privacy uh confidentiality, but also have a reporting obligations and reporting uh or or the the obligation to respond to regulators uh when the need calls for uh for some level of uh reporting. Um so it it's a hard balance to uh to balance but it is something that we think deep and hard about and we basically set the framework that's in a way similar to traditional finance. You want certain transactions especially your own payments and uh institutional level payments to be generally private.
At the same time uh you don't want to facilitate uh illicit transactions that are uh you know that that's just harmful for society and uh so we are also working on uh our privacy which is uh also taking a de approach uh on making transactions private.
>> Yeah. Okay. And and I we I think just through all the conversations I've had today with, you know, several great panelists, it it is becoming evident, you know, how important privacy is. Uh and this is something that spans, you know, both the consumer side, the merchant side, but also institutional retail on several different dimensions.
Uh several dimensions have, you know, they hold privacy to a high regard. Um uh one of the things that came up I I asked a previous panel about you know what is the what are some of the largest bottlenecks some of the some of the most limiting factors that we currently have in the agent economy to ultimately accelerate on this path. Um, one of one of the the bottlenecks that was brought up was uh around um chargebacks essentially, you know, there is going to be some early growing pains with, you know, agents going out there making transactions if we want to set these things loose. They're going to learn a lot, you know, but they're going to have to they're going to have to fail, make mistakes in order to do that, just like us. And so, you know, when they make something that is, you know, has to get returned or charged back or something like this, if we're using stable coins for this, it's kind of like you're kind of so, you know, like you kind of send the the crypto, it ends up in the other wallet.
There's really no like, you know, clawback mechanism uh for that. Um Gordon, have I I take it you guys are probably thinking through this, you know, big proponents of seeing this agentic economy take off. you know, how are you thinking through what agentic commerce looks like in practice?
You know, clawback mechanisms, these sorts of things. Um, as as we move this forward, >> it's a great question. So first of all I want to separate the settlement layer which is about you know settlement finality making sure the transactions actually going through versus the application layers on top which could include escro services could include you know uh uh dispute mechanisms those could be in the form of smart contracts uh along with potentially third parties uh that would allow things like chargeback uh things in the escro type of nature and but that's our building on top of a very solid foundation of uh finalized settlement. uh so those two are are distinct and we have done work on both fronts on both strengthening the settlement finality of USDC but also at the same time uh some work on escrow uh smart contracts um uh you know how how do you actually enable this um I think at the end of the day uh AI agents if they are operating in the real world handling large values of transfers not just nano payments they need to basically have uh a combination of identity and liability, right? Uh identity meaning that you can identify even if you cannot issue an ID to a agent per se, but having some way of tracing it back to uh someone that ultimately is responsible and liability as a way of um being able to provide that uh some some way of calling back, right? If AI agent misbehaves, perhaps you can slash a pool of asset that AI agent holds. Not so dissimilar to how uh how things work in the uh in real life with humans.
>> Yep. Yeah, it it makes sense. And I'm I'm excited to see some of these different uh design approaches uh take shape. you know, some of this one of the the greatest things about this industry I love so much is, you know, we can see experimentation in the open, you know, for all to see. Uh we can all learn from each other's mistakes and and ultimately build a better uh industry that way.
Guys, I I know we're running up on time.
La last question I have for you uh is around essentially just tracking this to understand how far along on this journey we are to see are we accelerating are we making the right design choices um what metrics should we be tracking tracking I you know I obviously on stable coins I know you know we've got you know the entire stable coin industry you know circle makes up a very significant part of that uh I think we're some somewhere around like 300 billion in total stable coins in circulation right Now, um, we talked about X42, you know, closing in, I believe, on a million, but 20 million transactions. Um, what are some of the metrics that you're tracking just to understand where we where the agentic economy is at, you know, it because if we're just tracking stable coins, that's so broad, probably doesn't capture a gentic economy, you know, extremely well. You know, we could be tracking X42, but that's probably too narrow.
What are some of the metrics that we should be tracking to really get a sense of how far along the Agentic economy is?
and and and you know that way we can continue to keep tabs on progress here.
>> Uh a spicy perhaps counterintuitive take here Robbie which is I think all quantitative metrics that we we can track is largely backwards looking >> the most interesting data that I'm looking for is qualitative intuitive data which is more forwardlooking. So all three of us are at the intersection of doing a lot of work in both crypto and AI. All three of us also span both enterprise and consumers. How much of our day-to-day call it commercial transactions do we have currently handled by agents? I suspect very little. So when we get to the point where we see actually real important problems that Agentic commercial transactions are solving for us, that is the most important intuitive data point that we can see that's forward-looking.
And after we see that, we will eventually see those numbers show up in backward-looking transaction types.
That's very well put. I I think uh they I absolutely agree with that. I think there's a lot of AI assisted transactions. You know, think of the last time you bought something because you took a AI suggestion and there's a conversion of how do you actually convert that to a full agentic type of payment. uh but I think there are some numbers you know as we go along we could look at right I've been looking at the percentage of transactions that is mediated by smart contracts as a broad proxy but a lot of those are defy contracts uh but you know it's something around 70% of transactions are mediated by smart contracts already right again if you take a aentic to be a broader sense of uh programmable and smart contract mediated and maybe some of the agents are doing defy transactions.
Another metric I've been looking at is uh the amount of small transactions under a dollar. Uh that's also at least provides some signals about not just adoption X42 but just the value of small small transactions how they add up.
>> Yep. Um I I just thought of one more and I I feel like I'd be remiss if I if I didn't ask. And um you know we we've are continuing to see a Gentic you know commerce explode. Um, I I want to ask about how stable coins are are relative to, you know, credit cards that also may be adopting, you know, agentic workflows because I know earlier we kind of talked about credit card processing fees and bringing those down and maybe stable coins can help with that. And we see like the likes of Visa and Mastercard, you know, start to implement a stable coin strategy, but also start to implement an agentic strategy.
What d I guess this is mostly directed at Gordon, but George, you know, I'm I'm curious for your thoughts, too. Like what direction do you guys do you think these these guys go? Like Visa, Mastercard, are they going, you know, deeper one direction or the other? How do you think this progresses?
>> I I think the card networks, if you think about what really draw people's to it is their specific incentive structure for essentially the uh interchange fees and kick back to consumers in the form of rewards, right? It's a very specific well-designed mechanism incentive structure that draws in the users that are reward sensitive and kind of uh the merchants are less reward sensitive and much more or less uh uh cost sensitive and much more sensitive to adoption. Um on the stable coin side you can think of as broadly generic programmable assets right you can make those same incentives work but you can also program a number of other well thoughtout incentives and test out different type of incentive and mechanisms I think that would drive adoption right it's not just one single set of rules but rather you can optimize tailor it to your merchant need design the best incentive structure that works you know for the specific use case and that's what's more powerful about it.
>> Yeah, I plus one to everything Gordon said. I think that there's a lot of advantages of essentially what we have in the broader stable coin world, but the probably the two or three most powerful advantages. One is is the ability to do structurally lower fees and second is just programmability. You can do programmability in traditional finance, the traditional payment world, but just so much harder. Like bringing the actual difficulty of creating programmatic finance, programmatic payments, um bringing the difficulty down by say an order of magnitude just unlocks so much more potential.
>> Yeah. Yeah. Well said. Um and uh I'm sure you guys saw this, you know, Stripe advent bid to take PayPal. I'm just cur Do you guys have any thoughts on this?
Any reactions? you know what is this it net positive? Do you think it goes through? What does it mean for the the future of payments if it does go through? What do you guys think?
Yeah, it's uh I haven't thought too much about it but you know it's certainly there is value to be had uh in uh their respective businesses and uh uh I think the the everyone recognize the value of the payment system and the network effect you can establish and this is why it's so important to have a uh you know both well- reggulated product but one has the depth and liquidity that's needed to enable broad adoption.
>> Yeah, I don't have too much context behind this either, Robbie, but I presume that part of this is just the broader consolidation happening in parts of the industry right now.
>> Yeah, makes sense. guys. Appreciate you uh uh taking the time uh staying a little bit longer uh with me answering answering the good uh the questions here and uh you know dude putting putting fighting the good fight uh put in the good work. Uh you guys are absolutely crushing on both sides. I mean circle Goliath and stable coins near uh you know obviously on the AI front as well as the intense front absolutely crushing on uh on all fronts there. Um, hope to see you guys again soon and uh, yeah, appreciate you guys taking the time and uh, wrapping up virtual near day.
>> Thank you.
>> We'll see you next time. Thanks, guys.
There you have it. That wraps us up. Uh, Circle and George from Near AI, Near Intense. Um, I mean, what a show, guys. I'm I'm zoned. I'm a bit cooked. Uh what a I mean just several great conversations. I feel like I learned a ton about the Aentic economy where we are uh right now. Appreciate you guys rocking with us. Uh we've got I mean just great combos, great asks uh great questions in the chat. Jordan regulation liquidity network depth. Yeah, that covers about every buzz word except actually telling us anything. Man, I guess you weren't watching the rest of the stream, buddy.
Uh Bashtick from Near Legion. Uh, Alfredo, can another token be used apart from stable coins? Absolutely. I think we're going to see Near Intense be a vehicle that can use pretty much any token. If the the merchant, you know, the seller wants stable coins, they can get it. The buyer can use whatever token they want. You just swap it uh on the way. Stable coins is a programmable incentive layer. It's actually a genuinely good frame. But the reason Interchange works is decades of habit and locked in reward psychology. You don't just reprogram consumer behavior by making the asset flexible. Completely agree. Had someone early early early in the rollup's history tell us consumer behavior takes a long time to change. It really doesn't change very easily. It's better to just keep consumer behavior the same and implement a cheaper, better, faster solution while maintaining consumer behavior. Shout out ox Luna. Shout out Sapper. sell private AI is near AI. Visa and Mastercard have been implementing stable coin strategies since like 2021. Still waiting for that to mean anything at scale. Completely agree. But you know what? It's uh it's slowly then all at once. And that that's what we're seeing play out in real time.
Bashtick, how will the agents actions be monitored? Yeah, that's what we're wondering ourselves because these harnesses are not really doing the trick. We need a way of keeping a human in the loop, but ultimately we just need a way of these things just doing a better job. Like I just want the agent to to do a good job, make the right decisions. And ideally, I don't have to be in the loop. That's why I use an agent. Uh Masa says slashing an AI agent's assets as punishment is a fun idea until you realize the agent doesn't care. The human behind it already moved the funds. Unless the funds are locked in a smart you know an escro contract or a smart contract staking contract in order you know to uh secure that is the principle of economic security. If you can have you know uh a regulated stable coin or some sort of uh coin collateral asset in the staking contract with some decent market cap that thing's not going to just disappear. I mean you're going to have it locked. Uh so you could slash it uh and in theory you could use that as a training mechanism uh for the AI uh which would be extremely cool guys. I mean you guys have been rocking with us been at it for almost three hours now.
Ox Luna, Flippa 01, Emit Jethro, Joshua, Abzar, Blake, Danzy, Tuma, Jpor, Badger, Marcus, Web 3. Guys, we are live every day, Monday through Friday. I see a lot of new names in the chat, so appreciate you guys rocking with us. Um, and so, uh, yeah, slashing an AI agent that doesn't care, bro. the agent who has no feelings. Only the agent who already yokeked the funds cares. They can't yoink the funds if it's locked in a smart contract. Guys, come on. Uh, King's Clips, you're an OG.
Appreciate you rocking with us today, guys. Tune in every Monday through Friday around noon Eastern, maybe around 12:30 or 1. We go live every single day to bring you guys the best and the brightest. We're going to keep you guys, you know, here. We got the easy job.
We're just asking the questions. The people that come on the show are fighting the good fight. they're doing the hard work and uh implementing these things in the builder trenches day in and day out. So come back, learn from the best and uh continuing to tune in.
Uh guys, I think the market is pretty much flat. So especially on a day like today when market's kind of flat, a little bit down, not really too many movements, don't need to, you know, you know, stare at the charts on a day like today. tune in, learn a thing or two, and uh ultimately use it to your advantage so that you can make better decisions in the market and beyond. Uh we will see you guys tomorrow, same time tomorrow. Uh and we're going to continue with another fantastic show to wrap up the week, guys. Thanks for watching.
We'll see you again tomorrow.
This has been the rollup. Peace.
[music] >> [music]
Related Videos
27:20
Blockchain & Crypto
Multi Vendor Multisig w/ Seed Signer, Hodl Dee & QnA
BitcoinMagazine
985 views•2024-09-05
16:20
Blockchain & Crypto
Oasis Week in Review: Latest blog articles, workshops and more
OasisFoundation
135 views•2024-10-18
9:29
Blockchain & Crypto
Kaspa: How ZK Turn Blockchains Into Settlement Layers (Part II)
cxc
1K views•2025-12-19
14:40
Blockchain & Crypto
以言會友 EP13｜當比特幣屢破紀錄 區塊鏈技術能帶來什麼？
dotdotnews
293K views•2021-01-05
1:33
Blockchain & Crypto
Soroban Development: Ecosystem Growth, and the Rise of 70+ Smart Contract Projects
SorobanOfficial
1K views•2023-07-19
1:15:25
Blockchain & Crypto
Balaji Srinivasan I The Fiat Crisis | Pragma Tokyo 2023
ETHGlobal
37K views•2023-05-06
11:09
Blockchain & Crypto
$22 million NFT scammers arrested (insider evidence)
coffeezillaextras
806K views•2025-02-03
1:01:22
Blockchain & Crypto
SYMMETRICAL TRIANGLE HOLDS THE KEY TO NEXT MOVE" DON'T IGNORE
xrpfuturemillionaire
800 views•2026-03-15
Trending
13:38
Law
WOW! Judge TURNS THE TABLES on Trump in His OWN $10B LAWSUIT!!!
MeidasTouch
197K views•2026-07-23
13:25
Business
Playstation NO DISC/NO BUY Fight Is Over...
DavidJaffeGames
4K views•2026-07-23
22:40
Business
Steam and Xbox Just Dropped The Hammer On PlayStation
OhNoItsAlexx
9K views•2026-07-23
16:52
Anthropology
Americans Confused in Australia for 17 Minutes Straight
IWrocker
17K views•2026-07-23