Author: Haotian
After recently chatting with some entrepreneurs and VCs, there is a common feeling that everyone remains confident about the AI + Crypto track, but is 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) Using MEME for Token issuance is no longer an advantage for AI Agents, and even "talking about tokens becomes taboo". If a project lacks PMF support and only has a Tokenomics cycle, it will naturally be labeled as pure MEME speculation, just a wolf in sheep's clothing, with little relation to AI;
2) The original landing order of AI Agent > AI Framework > AI Platform > AI DePIN may be adjusted. When the Agent market bubble bursts, Agent becomes a "carrier" after large model fine-tuning and data algorithms are formed. Without core technical support, it will be difficult for an AI Agent to show its muscles;
3) Some projects originally providing AI data, computing power, and algorithm services will 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-differentiated directions. Web2 Agents focus on Utility, so low-cost deployment development platforms work, but web3 Agents focus on Tokenomics. Overemphasizing low-cost deployment will only trigger 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 "fat protocol, thin application" logic. But how should the protocol be fattened? How to mobilize idle computing resources, drive algorithm low-cost application advantages through distributed architecture, and activate more vertical subdivided scenarios in finance, medicine, education? As for applications, how should they be thinned? AI Agents' autonomous asset management, autonomous intent trading, and autonomous multi-modal interaction are not achieved overnight. We cannot try to bite off more than we can chew. Demands must be subdivided and gradually implemented, otherwise, even a mature standard for a DeFi scenario would take one or two years;
6) MCP protocols and Manus automation in the web2 domain provide inspiration for innovation in the web3 domain. Directly extending development based on MCP + Manus for web3 application scenarios, or using distributed collaborative frameworks to enhance business scenarios above MCP, is feasible. 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 AI technology development.
