Decentralized AI & The Agentic Economy

A recent panel discussion on decentralized AI and the emerging "agentic economy" brought together speakers from Allora, Sage/Herd, Octane Security, EZKL, Gaia, Zo and Coinbase. Together, explore the potential of AI agents to revolutionize on-chain interactions, discuss the technical and economic challenges facing the field, and consider the implications for developers and users. Here is a summary of the key points of the discussion.

AI Agents & Mass Adoption

The discussion begins by exploring the potential of AI agents to drive mass adoption of crypto. One panelist argues that we could see a billion agents interacting on-chain before we reach a billion humans regularly using crypto. This is because agents can abstract away the complexity of managing crypto assets, just like financial advisors manage traditional investments. This could lead to a future where users simply give their agents instructions and allow them to manage their portfolios autonomously.

"You can just have a general idea that it's making a crypto portfolio for you...in the same way you just give money to a financial adviser and you don't have to think about it again," one panelist explains.

Another panelist compares the development of AI agents to the rise of TikTok, suggesting that agents will become adept at predicting and executing actions based on user interactions, creating personalized experiences. This vision positions agents as extensions of human intelligence, enabling users to offload tasks and navigate the complexities of the on-chain world more efficiently. This panelist views these personal agents "as an exocortex with our neocortex." I find this future very plausible. Generation Alpha already uses smartphones and tablets as extra limbs it’s fascinating to observe so over time with the introduction of more functional AI agents I see this future becoming true earlier than we think.

Multi-Chain Future & Parallel Execution Environments

The conversation turns to the technical aspects of the agentic economy, specifically the role of multi-chain environments. Panelists generally agree that agents will operate across multiple blockchains. Their ability to bridge between chains autonomously and without time constraints makes a multi-chain future likely.

"What's nice about the agents is that they don't care about the complexities of bridging nor are they time sensitive about it," one panelist points out. "If you just give an agent money, you give it a task and you know it's eventually going to get done. It could spend however long it wants to bridge to wherever to ultimately achieve your goal."

Parallel execution environments will be crucial for the scalability of agent networks, given the complexity and recursive nature of multi-agent systems. Blockchains like Monad, Solana, and Sui as well-suited for these applications.

However, one should caution that providing context to agents across various chains remains challenging, highlighting the need for robust data infrastructure and tooling on every blockchain. Without this context, agents will struggle to operate effectively in a multi-chain world.

If provided the right context  the potential for agents to understand how any chain works is high but can be a lot of work from a developer perspective.

Critical Missing Primitives: Payments & Security

The discussion then explores crucial technical primitives necessary for the agentic economy. There was focus on payments and security as key areas requiring further development. Regarding payments, the need for seamless payment rails between agents and humans is highlighted. Existing solutions like MetaMask's delegation toolkit, Agent Kit, Coinbase SDK, and Lit Protocol are promising, but merchant adoption remains a hurdle. Panelists expressed concern that established players like Stripe could dominate the space by creating closed ecosystems that limit interoperability. It’s also important to take note that there is an AI development boom because a lot of things have been open-sourced recently. Many billions of dollars have been poured into AI development. So, although there’s a staunch rise in development, it’s not entirely backed by loads of cash and experts as previous AI research has been.

Security concerns are also raised, with panelists highlighting the potential vulnerabilities of having agents manage on-chain wallets. Adversarial attacks on AI models and the underlying infrastructure pose significant risks. The use of zero-knowledge proofs (snarks and starks) to verify the correct execution of agent actions is a potential mitigation strategy. Still, the scalability of these technologies remains a challenge.

One panelist expresses concern about "introducing additional security vulnerabilities," describing it as "sort of an oracle setup where there's this thing potentially running off-chain that has access to a wallet and is then issuing transactions." Middleware is notorious for being the access point to many hacks. The panelists warn of potential risks, such as the introduction of adversarial examples and direct hacks to the infrastructure running the models.

Developer Experience, Memecoins, & the Future of the Agentic Economy

Lastly, the session concludes by examining the role of developers and the broader trends shaping the agentic economy. It is acknowledged that the rapid emergence of agent frameworks and the current dominance of memecoins are driving attention to the space. They emphasize the importance of providing developers with the tools and infrastructure needed to create compelling agent applications. Mobile-first experiences are crucial for capturing user attention and building direct consumer relationships.

"I think if you assume everything gets monetized, attention is the final value of currency," one panelist argues. "This [mobile] is the most dominant medium through which attention is valued. So one would assume if you're building an AI agent app you are trying to get into this and building for a consumer relationship” There are concerns about the sustainability of the current memecoin-driven hype and advocate for agents that provide real utility and value beyond speculation.

"I think the AI meme coins are bad," one panelist states. "Uh, and we probably shouldn't be using them as like the poster for the industry. Um, it's kind of an indictment actually of the the confluence of the two."

Looking ahead, they predict that agent swarms, DAOS, and user-owned agents will play significant roles in the evolution of the agentic economy.

Conclusion

The panel discussion provided a comprehensive overview of the opportunities and challenges facing the decentralized AI and agentic economy space. I do think the takes were a bit tame and if you’ve been following the AI x Crypto space for the past few months on CT you can come to these conclusions yourself but it was nice to have all the conversations in one place, sort of like a mini Crypto x AI state of the union. I am very curious about what predictions will come to fruition and what areas will fizzle out within the next few years. This is inspiring to see because while acknowledging the current limitations and potential risks, the panelists expressed optimism about the transformative potential of AI agents to revolutionize on-chain interactions, empower users, and create new economic paradigms.

Have thoughts or feedback? Join Chainflow on Discord or follow us on Twitter/X!