Cambrian Network CEO: Why AI Agents Are Taking Over DeFi

Added: 2026-05-12

[00:00:00]We are back from the Capital Summit here at the Ritz Carlton in Miami, Florida.

[00:00:05]Uh this is our AI show, AI Super Cycle.

[00:00:09]I'm here with Sam Green. AI Super Cycle is powered by Near. Uh they are doing tremendous things with Ironclaw, Near Intense, all the likes. And we're going to be at a near event on Thursday. But now we're here with Sam Green, the founder of Cambrian Network, and we're going to be talking about Agentic Finance, Agentic Commerce. You recently did an entire agent finance report and so we're going to be talking a lot about this. Uh Sam, how you doing, man? How you find in Miami?

[00:00:37]>> I'm doing great. This is my first time in Miami. Uh I've been listening to the show the past few days and it's been excellent.

[00:00:43]>> Appreciate it. Uh we are on the home stretch here. We did a full slate yesterday at the tokenized capital summit. Uh and we're rounding out on uh the home stretch here at the capital summit. Um, agents has been one of two conversations primary fields of interest here at the conference. It's been agents and it's been tokenization. And so I want to get a sense of this agent FI report that you put out there for Q1 2026. You tested a bunch of agents. What did you find when you were testing these agents and what did you put into this report?

[00:01:19]>> Yeah, so the the uh agent fi landscape is super interesting right now. Um it's about a year old with uh serious efforts >> and uh the the Q1 report was our latest report. We did our first one last year and we're going to have another one coming out in a month or so. So updating updating it and um I think what's interesting if we like maybe a little bit of context >> please.

[00:01:40]>> Ethereum went live 2015.

[00:01:43]>> A unis swap they went live about eight years ago. Mhm.

[00:01:47]>> And now about a year ago, we started to see these uh new new layer built on top of the existing D5 primitives.

[00:01:55]>> And um yes, so my team we tried about a hundred of the projects. So, it's a very it's a very popular space >> and we have we filtered out basically the 40 of the most of the projects that are most mature that are shipping that are that have users and um I can go into I can break down what the different types are if if >> that would be fantastic. Yeah, let's let's >> so what what we're seeing that had uh the initial the initial segment that had the most PMF was uh yield agents. Mhm.

[00:02:26]>> So these agents would basically look on chain, look for yield opportunities monitoring let's say all the vaults >> um and then basically depo deploy their users capital across all these vaults and then when API APYs change or became just changed they would rebalance and so many of the these are are rebalancing multiple times a day trying to get their users the best yield also trying to protect their users from risk. Uh for example, with what happened with the Kelpdow. Yeah.

[00:02:54]>> Um several of the companies that we work with uh they proactively withdrew their users funds before they were locked within a >> Oh, that's incredible. And they did this with AI agents.

[00:03:04]>> Yes. So maybe let's let's break this down and then we'll come back to the rest of the that market uh segment. So uh so what what we saw was that last year most of the agents that were getting PMF that were getting uh starting to manage their users capital they were rule-based. They were what we would call bots and then over time over the year what's happened is they're using more and more AI in their stack.

[00:03:30]So I'll I'll highlight Sale sale is a yield agent. It's a company that provides yield agents and u uh what they now I'm going to explain how they're using both like the rule-based logic and AI in their decision-m >> okay >> so what what they're doing is they have a they have a risk engine that they use and for the deterministic rule-based logic they're looking at how much yield is being distributed uh across all the vaults what is the what is the what are the APIs available and then [clears throat] they're also looking at volatility of the assets they're looking at if there's capital fight from the vaults and they're looking at uh Twitter. They're looking at crypto Twitter sentiment and this is all within their autonomous risk engine that sits on top of their uh rebalancing engine.

[00:04:16]>> And so you can see that it's basically a blend of rule-based logic and AI to be able to uh automatically uh rebalance and protect rebalance portfolios and protect their their users portfolios.

[00:04:30]>> Okay. And so I expect to see I mean I expect to see this combination of things a lot in the coming years both in Tradfi and in crypto where you have deterministic optimization methods that you use for you know well-known well-known things. Um and then you're going to be using AI to let's say look at the news see if there's something that could destabilize the market coming up and that can help basically uh protect users.

[00:04:57]>> Yeah. Incredible. So the primary use case for these agents right now are these yield uh these yield agents as you describe them. Uh and their primary purpose is to protect and uh make more efficient or optimize the yield that they're getting across this entire vault complex.

[00:05:16]>> Yeah, exactly. So what we saw, you know, as we entered the bare market, a lot of people went to stable coins. Yeah.

[00:05:21]People still want to get yield on their stable coins. And so that was one of the reasons these yield agents got a lot of traction was that users were basically just trying to get their best yield in their in their stable coins.

[00:05:33]>> Um the other the other segment that has been really popular are the trading agents. Okay.

[00:05:39]>> So these are agents of course that are these agents are are really heavily using AI >> and they're looking to look for opportunities, look for buy, sell signals, analyze uh analyze reports.

[00:05:52]Mhm.

[00:05:52]>> Um, so these are Yeah. So that's the other other segment. And then finally, just going through the other other types, we also have betting uh and prediction market agents, >> which this is this has had the least traction so far, which I'm surprised >> about.

[00:06:06]>> So we heard a lot about this on Twitter, but on on the ground, there really hasn't been a lot of development.

[00:06:11]>> That's right.

[00:06:12]>> Yeah. Not not much activity yet. Uh I expect to see more uh on this.

[00:06:16]>> Um we also have news and analysis agents. I don't know if you know like what AIXT is.

[00:06:22]>> Yeah, of course. Um I mean that that's one of the ones that has stuck around from you know the the original cohort of crypto AI apps >> and this is you know AI expertise this prolific tweeter uh that does some pretty good analysis. Yeah. Uh on on on Twitter.

[00:06:38]>> Yeah it's it's great. U I think um yeah so we have a handful of those.

[00:06:42]>> Um so that's that's a pretty good sub survey. you we have yield, we have trading, we have prediction market agents, and we have news news and financial uh uh analysis agents.

[00:06:53]>> And and you know, you also mentioned how there's rulesbased um uh bots as you call them and these are a little bit different than the agents themselves because they're not really embedding AI into uh their protocols. Can you how do how do you distinguish between bots and agents and what happens when we start to upgrade a lot of these you know simple algorithms into fullyfledged autonomous agents.

[00:07:23]>> Yeah. So I uh may maybe just to clarify most everything these days has an AI component and is using AI somewhere in their stack if not if if only for risk >> and analyzing what's happening outside of the block of blockchains. Mhm.

[00:07:38]>> Um but um maybe so maybe going into this, what is the definition of an agent and when does a bot become an agent? So I think it's it's helpful to clarify or to distinguish between agents >> and AI agents and I would say that uh you you know agents have software agents have been around for a long time. I would classify bots as agents. They're not AI agents.

[00:08:01]>> Got it. Um, and then what we're going to see across the board both in trifi and in crypto is we're going to be seeing what's going to end up I think happening is that there are things there are guard rails that basically the the proven techniques of of what bots use which was really like mathematical optimization.

[00:08:22]>> Uh there there are basically proven techniques that AI agents are going to use as tools as screwdrivers, hammers and whatnot. And so in the future, I expect to see more and more AI agents capturing users intent.

[00:08:36]>> And then these AI agents will have guardrails. They'll have a toolbox of little bots >> that they can then call on and execute.

[00:08:43]>> Yeah. Yeah. So they'll have sub agents, if you will, or tools that they can call on that ultimately is going to allow them to complete their mission, uh, which is, you know, completing and accomplishing the prompt that we set out for them.

[00:08:54]>> Yes. Um what of this is and you mentioned yield um yield agents as as one of the primary segments for this space right now.

[00:09:06]>> It it still feels like there's a lot of hocus pocus in this in this you know agent world as a whole. Um you know on the on the news and analysis we have AI XPTt doing well. We have some yield agents doing well. For someone that's looking to to start to take advantage and and capitalize on the opportunities um with some of these agents, where would you recommend that they get started so that they can start to navigate and avoid some of the pitfalls with these agents that are a little bit less effective um and and really start to flourish? What are some of the most um some of the agents that are able to provide the most value right now?

[00:09:43]>> Yeah. So, yeah, maybe going back to your earlier comment about the hocus pocus.

[00:09:47]Yeah, >> I think it's interesting because we're in a we're in this uh hobbyist era >> uh where people can download OpenClaw and they can spin up something on their own, right? They can connect a trading API and they can start trading. Um but then you also see things happening like uh you see things see happening like API keys being leaked.

[00:10:08]>> Yeah.

[00:10:09]>> Money, the thing doesn't do what you want it to do. So I think I think it's important to distinguish between the the hobby level activities enthusiasts right people who are semi-technical they can get their hands dirty they're doing this this is that's there's a lot of hocus pocus around that >> I think in terms of what is real if you I think a good place to go is to go to our blog at cambrien.org og and look at that report, that landscape document that we wrote and you could see that we've we've we've already basically vetted, we've tried all of the tools that are there. We think all of the teams that are uh that we have listed are credible and that their projects actually have user and they're managing capital. Um I think that's the safest that's the by far the safest way to get into this. Um yeah, is is through that.

[00:10:57]>> Yeah, absolutely. And so, you know, looking at Cambrian, uh, you've got an API into a lot of these agents. Um, and so maybe you could explain the API and then we'll also get into the road map of as far as how you see this world expanding and evolving.

[00:11:12]>> Yeah. So, at the heart of things, Cambrian, we we focus on financial intelligence. And so, what does that mean?

[00:11:18]>> So, specifically, we're monitoring onchain data and opportunities and risk >> and off-chain data. And we're basically collecting all of that data and we're organizing it and we provide it in a B it's a B2B product >> and we provide it to builders like these agent builders so that they can make more profitable and safer decisions.

[00:11:39]Yeah.

[00:11:39]>> And so for example in detail we what we do is we we're tracking things like uh a morpho compound unis swap all trading activity on chain >> um pricing on chain and we then go and dig deep we go get the details. So for example, if you right now if you wanted to get let's say the yield in A the easiest thing to do would go to would be to go to A's website and use their a API.

[00:12:06]>> We measure it uh directly from chain data and so specifically we have ported all the smart contract logic from all of these D5 protocols into our database.

[00:12:15]When we see transactions, we actually calculate how much yield is being generated, who made that, who made the money, what are all of the users positions, how much how much risk or how much leverage is there currently in the system.

[00:12:26]>> So, we're tracking all of that. We split it all out and then we provide it for uh we excuse me, we provide it through this API that's really easy to use.

[00:12:34]>> Yeah. And so I'd imagine that the the segment of yield agents could actually benefit quite a bit by plugging into this API because then they would be able to rebalance and optimize the efficient frontier of this yield more efficiently as a means of hey you know if I if I you know just saw the yield I'd think oh you know there's a reputable you know DeFi vault that's out there it's got a high yield. Let me switch over from what I'm currently in over there. But if you had some financial intelligence that you were able to pull directly from chain data that you could say, "Hey, this thing is actually overlevered, the the reason why is because they had this collateral asset that you know maybe moved off peg up or down. That's why the yield is spiking in this particular vault. Don't go in there. It's actually a a potentially adverse opportunity. You know, these these yield agents could prevent itself from making potentially poor decisions."

[00:13:24]>> Yeah, absolutely. So we're we're focused on tracking opportunity, the state of what's going on on chain, the the basically the low-level risk signals >> and providing that. And this is a lot of work. This is all we focus on. We've been working uh we were we've been working in this space for uh six years.

[00:13:42]We were previously many of our team was previously core devs within the graph protocol.

[00:13:46]>> Um and now we're hyperfocused on financial data and uh yeah. So our our goal is to provide this financial intelligence and then our customers then build products on top of these raw signals that now they don't have to run all this heavy duty infrastructure to have this comprehensive view of the market. Yeah. And and there are many different products that that can then be built on top of the raw signal >> and then maybe because of the I'll just I'll mention this because of the nature of this conference. Um we believe that the what as we know what's happening right now is uh within the last year we've had a lot of stable coins stable coin activity on chain.

[00:14:20]>> Yeah. And the next big topic is going to be stable coin yield.

[00:14:24]>> And institutions are going to be needing to the same information that we're providing to agents so that they can understand their opportunities, so that they can underwrite their activities, so that they can monitor risk, right?

[00:14:35]>> And um so we'll be expanding uh coverage into that. And like you said, there's hobbyists, there's enthusiasts. The these guys are deploying agents now and they're looking to integrate with all of the potential tools and analysis and intelligence that's out there right now.

[00:14:50]And then the institutions are going to come along and they're going to start to roll out their own agents both for their own internal access as well as from for their customer access. And then they're going to be looking at some of the intelligence and data products that that you're talking about right now selling to the enthusiasts and the hobbyists, but eventually selling to the to the institutions. What is it like building a product and I take it a data product?

[00:15:12]You probably, you know, there there's humans that could probably benefit from it as well that you know are making some of these decisions manually. But what is it like building a product that is primarily geared for agents? How do you >> you know like right now we're having a podcast. People are going to be listening to this. People are going to be listening to this. Like marketing for agents means different things.

[00:15:35]Eventually, maybe this this show makes its way into the agent verse. Like what is marketing, sales, product development? How do these things change when you go from a product for humans versus a product for agents?

[00:15:49]>> Yeah, great question. Um, maybe I I'd like to uh give a little background uh so on myself. So before I uh got into crypto, I did my my PhD was focused on AI, specifically reinforcement learning, which is the technique that has really unlocked a lot of the capabilities in LLMs in the past few years.

[00:16:09]>> So I've been thinking about agents and autonomous decision-m and safe autonomous decision-m for a long time.

[00:16:15]>> And so I was thrilled as soon as as soon as GPT came out and at the end of 2022, >> we started experimenting. And in 2023, we we created the first dashboard where someone could ask questions about blockchain data. We also did the first trade natural language trade where you could say I want to do this for that. M >> and and so in 20 we started this as a pilot project in my previous company these efforts and we started seeing uh oh the agents the LLMs were having a really difficult time for example writing complex SQL >> just like you know every time you use GPT it's going to give you a different answer and similarly if you want to write a complex SQL quer query for a database it will often be different and it will misinterpret or maybe interpret slightly differently your question each time and give you different results. M >> so one way that impacted us is that's why we decided from the beginning we were founded at the end of 2024 >> to create an an API as our first product >> an API that is very simple to use every everything is very very cleanly delineated and you can point an agent to one page and the and the agent can immediately know how to get the information without having to write anything. Um so that's that's kind of an that's an example of how we've tailored the product and it's also really easy for humans to to navigate. So you a human could be interested in yield. They can then go to a particular page and they can go see, you know, what is the the API call to get the yield data that they're interested in.

[00:17:38]>> Yeah. And and you know, you've done this by starting out with a ton of market research, you know, through this agent FI uh research report. Um next up, uh according to the road map here, you've got epoch one, the verifiability epoch.

[00:17:51]>> That's right.

[00:17:52]>> Uh followed by a mainet uh launch.

[00:17:54]That's right. uh take us through how you've uh you've you've segmented out these epochs.

[00:18:01]>> First we've got verifiability, then a mainet launch and and the mainet launch is where is the API live right now or or >> is it it's uh it's in uh private beta right now?

[00:18:11]>> Private beta right now. And so we've got verifiability to start then mainet then the convergence epoch. I'm very curious to hear about that and then the evolution epochs that follow. take us through verifiability, mainet launch, and convergence.

[00:18:25]>> Okay. Yeah, maybe I'll I'll tackle this from uh we'll look at uh uh chain link.

[00:18:31]>> Sure.

[00:18:31]>> Chain link uh they essentially provide pricing >> data, real-time pricing data.

[00:18:37]>> And the reason that they have become uh so popular is because they can provide it in a verifiable way. They have a track record of of providing it accurately, a track record of of high having high uptime. they don't go down and it's verifiable through their network of validators. Um similarly what we're providing is we're providing a comprehensive view of the onchain of the onchain market. We're providing a view of like we were talking about what are the what's the yield currently being uh currently available how much risk how much risk is there currently in the system and uh many other details. And so what we're going to be doing is we're our our current our when we go to market the initial product is going to be a centralized product that you'll pay with credit card >> but over time what we believe is that we can expand the DeFi design surface.

[00:19:29]>> Okay.

[00:19:30]>> By providing our data in a verifiable way which will come from a network.

[00:19:33]Yeah.

[00:19:34]>> And so you could think we're going to have an oracle network that can be providing yield opportunities the risk in the system user historical user activity. And >> you can kind of think of this another way to think of this is onchain the currently current onchain products like unis swap are very simple >> they just have the current price that's all they they have >> and if you think about a business a real world business what real world business only knows the current state of the business not the historical state they don't current uh businesses in the real world know what's happening at different chains >> that are that are around there uh their different locations none of that information is available for decision-making within DeFi today and we want to we want to expose expose that information on chain.

[00:20:16]>> Yeah. And increase the context window exactly um across you know several chains historical context all all of this. Um and so this is the verifiability epoch as you start to expose all of this data and make it available for both agent decision makers as well as as well as humans. Um and then the mainet launch um and so this will ultimately become launched. Um and you described this how you know you ultimately will be uh unveiling a lot of this data and a lot of this uh access to um you know this via you know these agents.

[00:20:50]>> That's right.

[00:20:50]>> Um and then you have the convergence epoch. I'm very curious about convergence. Been talking about the convergence for a long long time. You know both uh tradi and digital assets are converging. Uh and then ultimately finance as a whole is evolving into a more digital and agentic world. uh talk to us a bit about the convergence epoch and how you plan on scaling from there.

[00:21:12]>> Absolutely. So, just as a as a note, we we we built that website uh that website went live a year ago.

[00:21:19]>> And so, we've we we've believed in the convergence uh since, you know, for for quite a while.

[00:21:24]>> Yeah.

[00:21:24]>> And we we we see three things converging. I I heard your previous interview. I heard a note on the convergence of of AI and crypto. So that that is uh one area that we see converging. Um I think the the reason is is because as everyone knows operating on chain is complex especially when you're doing self-custody it's complex looking for opportunities on chain also complex and so we believe that the that agentic finance can lower the friction for for users >> and that's going to be done through automation and AI >> um >> and then of course we see a convergence there's a also a convergence with AI and finance >> right happening.

[00:22:07]>> Yeah.

[00:22:08]>> And then it's also clear now that there is a convergence between tradi and crypto is converging.

[00:22:14]>> Yeah. And so we we believe that blockchains are going to have especially when automation is layered on top >> uh through AI, blockchains are going to have the least risk, the least fees uh that that that uh it will have the best and it will have the best opportunities, >> most liquidity. Um, and so we think that that's where things are headed and that's where we're where we're trying to position ourselves is to be the the financial intelligence provider for that convergence. So we're starting with this small this emerging market of of agent pie.

[00:22:50]>> We know that as Trad starts to do onchain activities that they're going to need the same information for for their decision-m and for their underwriting, >> right? And then we also know that we have to build a product that can be easily used by AI because that's what developers are using to develop. So our tool has to be very trivial to integrate and then our the speed of decision-m is going to we're going to be seeing the speed of deci financial decision-m increase.

[00:23:15]>> And so we've also designed our whole system from scratch so that we can provide that information provide fast comprehensive and verifiable information >> for the for the this high-speed uh financial future.

[00:23:27]>> Yeah. Fascinating. Uh Sam, as we close out here, you know, any closing thoughts from your end? This has been a fantastic conversation on our AI super cycle show.

[00:23:37]Also, please, you know, tell people where they can find uh yourself and where they can uh learn a bit more about Cambrian. Um I'll tell you right now, it's cambrian.org or and uh and and where can they find you learn a little bit more and and maybe get in touch to ultimately uh utilize and capitalize on the financial intelligence that you're that you're talking about with respect to this convergence.

[00:23:57]>> Thank you. Uh yeah, so I'll I'll think I'll just leave with uh my Twitter handle is 0x Sam Green. That's that's a good starting point to follow what we're doing.

[00:24:08]>> Perfect. Sam, thanks so much for joining, man. Absolutely pleasure. Yeah.

[00:24:10]Enjoy the rest of your conference.

[00:24:12]>> Appreciate it. E2.