After recent conversations with entrepreneurs and VCs, there is a common feeling that people's expectations for the AI + Crypto track remain firm, but they are somewhat confused about the narrative evolution of Web3 AI Agents. What should be done? I have organized several potential directions for subsequent AI narratives for reference:
1) Issuing tokens through MEME is no longer an advantage for AI Agents, and even "talking about tokens is taboo". If a project lacks Product-Market Fit and only has a set of Tokenomics spinning, it will naturally be labeled as pure MEME speculation, merely a wolf in sheep's clothing, with little relation to AI;
2) The original landing sequence of Agent > AI Framework > AI Platform > AI DePIN may be adjusted. When the Agent market bubble bursts, Agents will become "carriers" after large model fine-tuning and data algorithms are formed. Without core technical support, it will be difficult for an AI Agent to flex its muscles;
3) Some projects originally providing AI data, computing power, and algorithm services may surpass AI Agents and become the focus of attention. Even if new AI Agents are launched, Agents created by these AI platform projects will be more market-convincing. After all, teams that can operate an AI platform are much more reliable than developers deploying at low cost based on a framework;
4) Web3 AI Agents can no longer directly compete with Web2 teams. They must find Web3's differentiated direction. Web2 Agents focus on Utility, so low-cost deployment development platforms work. However, Web3 Agents focus on Tokenomics, and over-emphasizing low-cost deployment will only create more asset issuance bubbles. Undoubtedly, Web3 AI Agents should innovate by combining blockchain distributed consensus architecture;
5) The biggest advantage of AI Agents is "application-first", following the logic of "fat protocol, thin application". But how should the protocol be "fat"? How to mobilize idle computing resources, drive algorithm low-cost application advantages through distributed architecture, and activate more vertical scenarios in finance, medicine, education? As for how applications should be "thin", autonomous asset management, autonomous intent trading, and autonomous multi-modal interaction are not achieved overnight. One cannot try to bite off more than can be chewed. Demands must be subdivided and gradually implemented, as even a mature standard for a DeFi scenario would take one or two years;
6) MCP protocols and Manus automation execution in the Web2 domain provide inspiration for innovation in the Web3 domain. Developing directly based on MCP + Manus for Web3 application scenarios, or using distributed collaborative frameworks to enhance business scenarios above MCP, is possible. There's no need to talk about overthrowing everything; being able to appropriately optimize existing product protocols and leverage Web3's irreplaceable differentiated advantages is sufficient. Whether Web2 or Web3, both are in this AI LLMs revolutionary process. The ideology doesn't matter; what's important is truly promoting the development of AI technology.
