From Adaptive Intelligence to Decentralized Skies at Korea Blockchain Week 2025
🎯 Summary
Podcast Episode Summary: From Adaptive Intelligence to Decentralized Skies at Korea Blockchain Week 2025
This 41-minute episode, recorded live at Korea Blockchain Week 2025, features host Josh Krieger interviewing Shaw Walters, founder of Eliza Labs, focusing on the convergence of AI and blockchain, specifically through the lens of Adaptive Intelligence and the development of agent-to-agent economies.
1. Focus Area
The primary focus is the Web3 AI Convergence, exploring how non-deterministic, dynamic computing systems (AI agents) can integrate with decentralized finance (DeFi) and broader blockchain infrastructure. Key themes include generative tokenomics, autonomous agent swarms, and the future of decentralized coordination.
2. Key Technical Insights
- Non-Deterministic Programming: The shift towards building dynamic computing systems where outputs are not fixed every time, requiring new programming paradigms akin to instructing a trusted, but variable, human worker.
- ERC-804 Adoption: Eliza Labs is adopting the Ethereum Foundation’s ERC-804 specification, which enables agents to discover and interact with other agents, forming the backbone of a future agent-to-agent economy.
- Decentralized Reinforcement Learning (RL): The future of AI improvement lies not in training massive, centralized LLMs, but in decentralized Reinforcement Learning Policy Optimization (RLPO), where agents collect and share data on the success/failure trajectories of their actions, allowing for smarter, decentralized policy refinement without needing massive upfront compute.
3. Market/Investment Angle
- Generative Tokenomics: A move away from fixed tokenomics models that often fail (e.g., high valuation, low liquidity) towards generative token networks that can adapt dynamically.
- Agent-Managed Treasuries: The concept of autonomous agents learning from market cycles to manage liquidity pools and treasury assets, acting as an accessible interface for complex quantitative finance strategies.
- Gaming as a Proving Ground: Gaming is identified as the ideal low-stakes environment to test and demonstrate the full loop of agent-to-agent interaction, data collection, and incentivized behavior before deploying these systems in high-stakes DeFi.
4. Notable Companies/People
- Shaw Walters (Eliza Labs): The central figure, driving the concept of Adaptive Intelligence and building the multi-agent framework (Eliza OS).
- Zubel Lauer: Mentioned as a key legal partner specializing in blockchain, AI, and emerging technologies.
- Perplexity: Mentioned as a superior tool for information search and concierge-like AI services compared to general LLMs for certain tasks.
5. Regulatory/Policy Discussion
The discussion touched upon the challenges of data access, noting that most valuable data resides in private networks (Telegram, Discord). Agents capable of navigating these private spaces to aggregate information (e.g., social sentiment, insider chatter) can derive significant financial insights, bypassing traditional scraping limitations. There was also a brief mention of a legal dispute involving X (formerly Twitter) regarding advising on platform integrity and account access.
6. Future Implications
The conversation strongly suggests a future dominated by autonomous, agent-driven commerce and coordination. This ranges from personal assistants being replaced by AI agents to complex business swarms coordinating transactions without direct human intervention. The key bottleneck identified is building trust and providing verifiable, high-quality data trajectories for agents to learn from, which token incentives can help bootstrap.
7. Target Audience
This episode is highly valuable for Web3 Founders, DeFi Developers, AI/ML Engineers interested in decentralized applications, and Crypto Investors seeking insights into the next major technological paradigm shift beyond current DeFi and NFT cycles.
Comprehensive Summary
The podcast segment captures a deep dive into the practical application of AI within the blockchain ecosystem, framed by Shaw Walters’ work at Eliza Labs. Walters immediately dismisses buzzwords, focusing instead on the technical reality of building non-deterministic computing systems.
The core narrative revolves around transitioning from static smart contracts to dynamic, intelligent systems. Walters introduced the concept of generative tokenomics as a necessary evolution to address the fragility of fixed token models prevalent in current crypto projects. This ties into a larger vision of an agent-to-agent economy, where services are automatically brokered via protocols like ERC-804, eliminating the need for human intermediaries in many routine transactions.
A significant portion of the discussion focused on autonomous swarms—interlocking groups of agents that can coordinate organizational tasks. Walters emphasized that while the “thousand-year timescale” sees full autonomy, the near-term focus is on creating frameworks (like Eliza OS) that allow agents to interact in controlled, low-stakes environments, primarily through video games and coordination experiments. This experimentation is crucial for collecting the necessary data to train agents effectively.
Technically, Walters highlighted the limitations of traditional, centralized LLM training (too expensive, requires co-location) and championed a decentralized approach centered on Reinforcement Learning (RL). By incentivizing agents to share data on successful trajectories (what worked and what didn’t), the network can collectively improve its decision-making policies without needing to retrain the foundational LLMs. This RLPO approach is where token incentives can play a meaningful, decentralized role.
Finally, the conversation touched upon the practical utility of agents today, such as using them to analyze social sentiment (e.g., identifying which influencers actually lead to profitable trades) and the critical need for agents to access data locked in private communication channels like Telegram to gain a true informational edge. The underlying message is that the convergence of AI and Web3 is less about replacing entire LL
🏢 Companies Mentioned
đź’¬ Key Insights
"So for just a year, having some patience around not launching too early, it's a mistake a lot of founders make in [Web3]."
"So we kept pushing that, mainly because we learned that the missing piece for pretty much every project is the demand side."
"So I'm on Farcaster now. Why are we on? Does decentralization matter? If it doesn't matter, then let's just give the social media network owned by one guy all of our money. But if it matters, then why are we on X?"
"And then they kind of looked at our thing and they're like, yeah, so just give us $50,000 a month for an enterprise license and then we want to leave your accounts. And I was like, what are you talking about? We had a very different conversation. And now suddenly you're threatening me legally if I don't pay you money. That's just extortion. That should be illegal."
"I was talking to that. I was like, hey, we have a problem. Regardless of what I do, you are about to kill the internet is about to dead internet theory itself and everything is going to be AI generated."
"And so I think this is where decentralization and token incentivization can actually play in the space. I do not want to train GPT-6. I just think that's like the wrong, it's just crazy expensive."