Podcast: The Round Trip
Compiled & edited by: Yuliya, PANews
When AI technology minimizes the cost of on-chain monitoring, a user's privacy could be instantly exposed simply by entering an address. The extreme transparency of blockchain faces a serious threat in the AI era.
In "The Round Trip" series of Founder's Talks, co-produced by PANews and Web3.com Ventures, host Cassidy Huang invited Kieran Dennis, an early contributor to the Core Foundation and co-founder of Z Protocol. Kieran is currently working on building Z Protocol, the first "Satoshi Plus" licensed blockchain— a blockchain specifically designed for scenarios integrating privacy and the AI Agent economy . (*Note: Satoshi Plus is a mechanism created by the Core Foundation that combines DPoW and DPoS to improve the security and performance of blockchain without sacrificing decentralization.)
In this dialogue, Kieran not only shared his experience in building Core DAO, but also explored why on-chain privacy has become more important than ever in the AI era, and how Z Protocol will build a secure, efficient and compliant decentralized operating framework for AI Agents through verifiable reasoning and zero-knowledge proofs.
The first Core-licensed blockchain, built specifically for the Agent privacy economy.
Cassidy: Welcome to this episode of Round Trip. Today we have with us Kieran Dennis, co-founder of Z Protocol . He is currently leading the development of the first "Satoshi Plus" licensed blockchain, a chain specifically designed for scenarios where privacy and the AI Agent economy converge. Kieran, it's great to have you on our podcast today. Let's get started. First, tell us about Core. What does Core do?
Kieran: I joined Core about four years ago and have been there for almost four years. Core is a Bitcoin scaling solution, and more recently it has been referred to as the “Bitcoin grid” because it is powering a range of revenue-generating applications focused on the Bitcoin space, and that value ultimately flows back to Core token holders.
One of the most important aspects of Core is that it supports some of Bitcoin's core use cases in terms of both yield generation and ledger infrastructure. Currently, the most interesting uses of Bitcoin are either for generating yield or for collateralized lending . Therefore, Core is primarily focused on exploring and advancing these use cases, while also considering how to become the infrastructure for more markets and expand opportunities within them.
On the yield side , Core is advancing Core Alpha, a Bitcoin yield platform for institutions;
On the collateral side , Core will partner with the consumer-facing wallet Satpay, allowing users to earn returns on their Bitcoin while simultaneously borrowing money, making the overall experience more suitable for ordinary users.
Another point that excites me the most is that Z is the first project in the Satoshi franchise chain that you just mentioned.
Cassidy: So what exactly is Z?
Kieran: Z is a programmable privacy platform for autonomous applications and anonymous activities. So, in a sense, Z is the purest expression of the Crypto + AI narrative. It's a completely new blockchain platform designed for a range of applications built specifically for AI agents.
This is precisely where Z and Core converge. We have partnered with Core, adopting their Satoshi Plus consensus mechanism and applying it to Zcash. This will allow Zcash miners and Zcash (ZEC) holders to participate in Z's consensus system. Simultaneously, we will leverage Core's technical capabilities as a Bitcoin grid to support the development of the Z Protocol and participate in revenue sharing opportunities, benefiting both Z users and the Core community.
On-chain privacy crisis in the AI era; blockchain is inherently suitable for agents.
Cassidy: That sounds really cool, especially the Zcash and Z Protocol part. Around late summer to early autumn last year, Zcash suddenly became incredibly popular, with many KOLs like Naval and Mert discussing it. What are your thoughts on the development of the Zcash community along this path? How will Z Protocol further promote the maturity of the Zcash ecosystem?
Kieran: I've always been a big fan of Zcash. Lately, Zcash has become very popular, not just within the crypto industry, but also in Silicon Valley's VC circles and among cypherpunk idealists.
ZCash's increased attention comes not only from Silicon Valley's growing awareness and popularity, but also from the ongoing technological development surrounding it.
For example, after Zashi (now Zodl Wallet) was launched, it set privacy pools as the default option, and subsequently privacy pools experienced unprecedented growth.
Then projects like Near Intents emerged, and Zcash's transaction volume increased significantly. I think it's one of the few crypto projects that can truly have a positive impact on the surrounding technological development.
Privacy is becoming increasingly unavoidable and important in the crypto world. From a human perspective, we have already witnessed the terrible consequences of extreme on-chain transparency, such as kidnappings and "wrench attacks" that use coercion to steal cryptocurrency. All of this essentially stems from overly transparent infrastructure, where once someone finds the slightest clue in your identity information, they can trace your entire digital life.
As we enter the AI era, this will become even more important. This is especially true in Silicon Valley, where the impact is becoming increasingly apparent. First, AI has virtually reduced the cost of mass surveillance to near zero . It's incredibly easy now; you simply give an address to a bot and say, "Hey, give me a complete research report, find out who this is, and who they've transacted with." If you're active enough on the blockchain, someone could potentially connect your entire digital life. At that point, there will essentially be no privacy left.
Cassidy: Yes, I was talking to Nansen AI recently, and they've already labeled over 500 million addresses. This confirms your point: AI is indeed significantly deanonymizing a large number of users, making it easier for outsiders to see who they are, when they did it, and what they did. Speaking of the design of on-chain privacy platforms, you cater to both human users and AI agents. How does Z Protocol balance privacy and compliance?
Kieran: Privacy is very important to humans, but even more so to agents. This is because agents transact and interact on-chain on a scale that humans simply cannot achieve. If they want to hide different strategies, counterparties, or identity profiles, it's impossible to do so within a transparent on-chain environment. Privacy is the first prerequisite for blockchain technology to achieve large-scale agent-based applications.
I built Z Protocol with the intention of enabling agents to utilize the blockchain framework. Blockchain is inherently well-suited for agents, but mainstream crypto networks are fundamentally not built around the use cases of agents.
Regarding the balance between privacy and compliance, this is especially important for high-net-worth individuals and institutions . They need to verify their activities and provide necessary transparency to their counterparts. Therefore, we will definitely provide some self-activated compliance features, allowing users to decide which on-chain activities need to be disclosed. This is entirely selective disclosure, actually closer to the practices in the traditional banking system than the highly transparent system of the crypto world.
On-chain reasoning, TEE, and ZK proofs
Cassidy: At the architectural level, for agents, if transactions such as fund transfers are to occur between them, a compliance mechanism like "Know Your Agent" and reputation scoring might be needed. How do I know that the agent won't exaggerate the inference results? How do you plan to incorporate these capabilities into the design of Z Protocol?
Kieran: Absolutely. A fundamental design feature of Z Protocol is to make inference a kind of on-chain primitive. Just like you call other smart contracts, you can also make an inference call.
Essentially, it involves sending the request into our privacy pool, which then collaborates with various models to route the request to the correct model. It then runs within a TEE (Trusted Execution Environment). This allows you to invoke the inference and retrieve the desired result.
It also provides verification, proving that it did indeed use the model you specified. Therefore, if you pay to invoke a top-level model, the entire process will remain private.
Cassidy: How do you know that you actually received a response from a top-tier model?
Kieran: The answer lies in the underlying infrastructure, mainly through ZK (zero-knowledge proof), and the result itself can verify this activity.
Additionally, as you just mentioned, you want to verify whether your agent is doing what you require. This system will also provide you with an inference receipt. This receipt will accurately tell you how the entire process was conducted, thus ensuring that the agent is indeed performing its duties.
This is actually a new approach, in a sense, approaching the truth on-chain. It also brings us back to the vision that everyone had when discussing smart contracts, such as "Code is Law"—if it's written into the code, the result should be the result you want. But in the past, we haven't really seen this hold true because there's always been a missing agent layer, or rather, a lack of such a verifiable proof mechanism.
And now, we can finally achieve this in the near future. This is truly exciting because it will unlock many things. For example, at some point in the future, it will be possible to achieve unsecured or low-secured lending if the agent has accumulated enough credit and can prove its on-chain history.
Cassidy: Regarding the ZK proof, who can raise questions? How difficult would it be?
Kieran: Anyone can use ZK proofs for auditing. It's even easier if you initiate the request yourself. You can automate it, using ZK proofs to determine whether the model is being used correctly within the TEE and whether data has been leaked.
In fact, it works like this: you can completely encrypt all input data and send it in privately, but at the same time, you can verify the model and that the TEE did indeed use the model you specified and followed the steps you required in advance. All of this can be verified.
Building Inference Networks and Model Markets
Cassidy: How do you plan to build this inference network? Many hyperscale cloud vendors are currently facing resource constraints. Will you design some kind of incentive to encourage more people to join?
Kieran: Yes. We've seen some very exciting paths. For example, Venice AI has made some progress in inference, although it's not fully encrypted or private yet, but we can build on that model to achieve greater inference availability . Another example is Bittensor , which has shown great enthusiasm in the market for model training, using token economics to incentivize participation in model training across different scenarios. These will all be key parts of Z's long-term roadmap.
Agents don't just participate in model training at the abstract level, training various different models; they also train themselves on how to use these models on Z. In this way, agents become increasingly familiar with how to use them, and may even eventually build their own blockchain and establish their own agent economy on Z.
So the "endgame" scenario you just described is actually about making reasoning capabilities truly available on a large scale, extending to those who previously had no access to it, including the agents active on digital infrastructure. These agents then train their models in the way they want. Once this training capability is mature enough, agents can start building the applications themselves. At that stage, things may already be beyond our current understanding; we may not know what the agents are actually trying to build— it could be an agent-to-agent lending protocol, or various other things.
Cassidy: It sounds like you'll most likely be doing some model fine-tuning first. After that, will you continue working on it, such as developing your own model? Or what's your overall strategy?
Kieran: To some extent, yes, but I think what's more exciting is creating a model marketplace where everyone can upload their models. For example, you could pledge a certain amount of assets to prove your model is compliant with regulations, and then there would be real competition on Z. It would see who can offer the best model, the best system, and the optimal input. I think that would generate the strongest network effect because the market itself is very powerful.
Different people want to train different models. Some want agents that excel at building on-chain applications, while others want agents that excel at trading and avoiding slippage. We don't want to make it a permissioned network; instead, we want various model providers to compete with each other on this platform.
Bottlenecks in the Agent Economy and Z Protocol's Solution
Cassidy: So, following your earlier assessment, when the agent economy truly begins to take shape, what major obstacles or bottlenecks do you think it will face next?
Kieran: There are two main problems:
The first issue is the lack of privacy. If privacy is not natively built into the protocol layer, it is far from sufficient for agents running on the blockchain.
Secondly, the framework is unfriendly . Current blockchain network frameworks are not agent-friendly. If you simply hand a wallet to an agent, it will likely engage in illusions while wasting a large amount of tokens. It's like handing over control of a computer to Claude, instead of Claude Code, which is specifically designed for execution. What you really need is Claude Code, which provides a framework for operation that makes the process more efficient, saves tokens, and reduces illusions.
Z's overall goal is to become that framework, enabling agents to easily perform operations on the blockchain. We've written access control directly into the protocol layer, allowing humans to explicitly tell it what it can and cannot do. We also provide clear skill documentation, LLM-oriented documentation, and a secure operating environment.
Furthermore, the Z ecosystem will be a vertically integrated application stack, all deployed on Z, and we will develop these ourselves first. Of course, external developers will continue development on the outer layers of the ecosystem. But all underlying primitives will be designed to be as composable and interoperable as possible. In this way, any agent that wants to trade on Z Trade, then establish a position around the traded asset, and further operate on Z Lend can complete these actions very smoothly. Moreover, there will be little illusion, and there will be no need to conduct due diligence on each protocol individually. For agents, this whole thing will essentially be like a meta-protocol, and the entire operation will serve them.
Cassidy: So this is somewhat like an agent harness? Like OpenClaw, Hermes, Oh My Code, etc., where you put in the key and it knows how to run it and can complete the operation within a certain framework.
Kieran: To be more precise, Z itself is the framework. You'll continue to use Claude Code or other applications to access the Z Protocol website and figure out how to perform the actions you want it to do. It will follow the most token-efficient and smoothest path. The entire process can be executed autonomously; you only need to give it the permissions you're willing to grant, and it will manage the entire on-chain life for you. In the future, the only interface we'll need will be the command line, which will truly drive blockchain towards mass adoption.
AI presents a significant opportunity for blockchain; now is the right time to invest.
Cassidy: Everyone's talking about AI and AI agents. So what do you think is the most likely narrative for practical adoption?
Kieran: The biggest opportunity in crypto history is right here. Blockchain tracks are designed for AI agents that need permissionless protocols, programmable execution tracks, and trusted, neutral ledgers, and don't want to rely on inefficient human systems.
It's not quite there yet because cross-protocol interaction remains difficult for agents, and the documentation isn't complete. But once you get past 0 to 1, going from 1 to 100 will be extremely easy. Once you give agents enough tools to think, act, and learn, they can continue building systems on top of the primitives we humans have constructed, truly accelerating on-chain adoption. And I think the truly interesting moment begins here. At that point, it won't just be about agent commerce; agents will genuinely build their own businesses on-chain, putting almost everything within these permissionless yet programmable protocols.
Cassidy: Releases like Opus 4.6 have greatly propelled OpenClaw forward. Do you think the time has come for a blockchain version of OpenClaw? How far are we from "going from 0 to 1"?
Kieran: The barrier to entry is actually quite high, and the reason is simple. You need to have a high enough level of trust in these things before you'll actually hand over your funds. Starting an OpenClaw Agent is easy, and getting it to post content is also easy, like posting on an agent-based social media platform. But if you actually give it even $100, things are completely different. So you have to build it in a way that users can trust most and that is most secure. This does take some time, but not too long, because this evolutionary path itself is very efficient.
I think we're very close to that moment when "infants grow up." This requires progress in two areas:
AI models must become smarter.
Blockchain technology must become simpler and easier to use . Deploying an agent on the chain now is like asking a two- or three-year-old to use a credit card.
We need a ladder of trust. If we want people to start using AI agents for crypto, there has to be a story: "Hey, my agent is making money." This may be the only way to drive mass adoption.
Cassidy: Given that we seem to be in a bear market right now and everyone is losing money, how is this going to happen?
Kieran: In my experience, bear markets are actually the best time to build and test technology . If we were in a crazy 2021-style bull market, adding the Crypto and AI narratives would be very dangerous. Because then, there would be too much frenzied behavior in the market, not just in token valuation, but also many people deploying things that are fundamentally insecure, and no one would really care about the underlying execution frameworks needed for AI to run on-chain. The end result could be this: once the entire system collapses, trust completely crumbles because the system wasn't even ready. It's better to let adoption grow naturally from the bottom up, step by step.
We are in the midst of a seismic shift in the technology sector, but it is not yet fully priced in. There are many reasons behind this, including geopolitics and many other factors. But ultimately, now is a great time to deeply invest in both AI and crypto simultaneously. The combined power of the two is immense, and they are far from becoming mainstream.
Adding to the factors I just mentioned, such as the existence of these digitally native orbital services, their TAM (market space) will be enormous, and their value capture potential will be huge. And frankly, I even think people don't necessarily need to profit through the old zero-sum game-style cryptocurrencies. Because I think people will start building truly interesting things, such as new consumer applications or financial primitives that have never existed before. These things can now be created on-chain and continuously generate revenue. That will be a true positive-sum economy, not the traditional PVP-style crypto mentality. I think we will see a real boom in crypto construction in the next year, and this wave will be truly driven by AI + crypto.
Technological trends are the long-term driving force behind cryptography.
Cassidy: Many traditional financial instruments (such as the S&P 500) are now entering the crypto world. This is more like a PVE (Proof-of-Life) approach; agents can use these tools, not necessarily for investing in cryptocurrencies. Furthermore, companies like OpenAI and Anthropic have already scraped content from across the internet, and users are constantly inputting new information. People will start to be wary of what information they are revealing to agents.
Kieran: Absolutely agree. Interacting with agents without privacy protection is extremely dangerous. Once people realize this, and further realize that there are opportunities to use very powerful models in a private way—all you have to do is use a protocol like Z—it will become an extremely powerful adoption mechanism.
It doesn't need to be driven by transactions at all; it's purely about "paying for reasoning." We've already seen the prototype in products like Venice. In the future, AI + crypto will unlock many different "building blocks" that can be combined layer by layer. Therefore, in the future, it may not be necessary to rely on PVP to attract people to crypto. I don't think we even need to wait for a long-term bull market; I believe that the long-term technological trend itself is the real driving force that will bring people back to the market.
Cassidy: Finally, what are you most looking forward to right now?
Kieran: I'm really looking forward to what the collision of Crypto and AI will produce. Privacy is our first step. Going forward, it will be like watching our own child grow up; we'll see agents become increasingly adept at using Crypto tracks, thinking, acting, and learning on-chain – that's how you build the strongest models.
Moreover, we don't have to sacrifice privacy, nor do we have to completely surrender ourselves to a surveillance-based state apparatus. This is the true way to put the cypherpunk ideals embodied by Crypto into practice, and in a way that truly embraces the AI era. This is not only an existential question for the crypto industry, but also crucial for society as a whole—can we solve the specific issues surrounding privacy and the new challenges brought about by the AI era?
