Several thoughts on the landing scenarios of Web3 AI Agents

Written by: Haotian

Further thought about the landing scenarios for web3 AI Agents, extracting several forward-looking insights as follows:

1. The most native application function of a web3 AI Agent may not be "trading." Although DeFi trading Agents have always been regarded as the Endgame form of Agents landing in Crypto, AI itself carries ambiguity reasoning and hallucination processes, which inherently contradict the precision and low tolerance for error required in trading scenarios.

In my opinion, the advantage of short-term web3 AI agents lies in the level of "data cleansing" and "intent parsing", rather than landing on the absolute accuracy of asset transaction execution layer all at once. For example: cleaning on-chain + off-chain applicability data to build an effective information graph; Another example: modeling and risk appetite analysis of on-chain user trading behavior, customizing Smart Money trading decision-making assistant, etc.;

2. The need for the web3 AI Agent for the A2A communication protocol functionality may be greater than that of MCP. This is because MCP calls are relatively mature functional API interfaces. If there is a mature Agent application ecosystem, MCP can perfectly solve the data island problem. Conversely, if the application business itself is not mature, the standardized interfaces of MCP will lack utility.

In contrast, the A2A protocol can create a certain incremental market for Agents, which will give rise to a batch of specialized vertical Agents, such as on-chain data analysis Agents, smart contract auditing Agents, MEV opportunity capturing Agents, and so on. The built-in Agent capability registry and P2P messaging network in A2A will encourage each vertical Agent to better adapt to the combination of value through interaction and complex interactions. If it only stays at the MCP protocol level, it is likely that web3 AI Agents will struggle to break through the limitations of language interaction.

3. The demand for infrastructure construction by web3 AI Agent > Application landing. In the context of web2 AI, pursuing the practical value of agents is naturally the highest priority. However, for web3 AI Agents to build a complete ecosystem, it is necessary to fill the severely lacking underlying infrastructure, including a unified data layer, Oracle layer, intent execution layer, decentralized consensus layer, etc.

Instead of clashing head-on with Web2 at the application layer (which is destined to incur losses), it is more appropriate to find a new path at the infrastructure layer and build an infrastructure with differentiated advantages of Web3. Although there is a relative lag in application implementation compared to Web2 AI, the urgency to build decentralized consensus networks for A2A operations and to create unified interactive operation standards for MCPs is not much less than that of application implementation, as these foundational infrastructures naturally align with the native characteristics of blockchain.

4. The shift in mindset from Crypto Native to AI Native. Looking back at the history of Crypto over the years, the mere adherence to the framework of "decentralization" has given rise to a rich variety of tracks and waves of innovation. In the future, the AI + Crypto field may go further around the concept of "AI autonomy."

Whether it is Agentic or Robotic, it essentially seeks a new AI-centered paradigm framework, such as a cluster of AI Agents with self-funding management capabilities, a template for smart contracts that can self-upgrade based on network environment and feedback, and a DAO governance framework that dynamically adjusts and optimizes based on community contributions. Ultimately, the key point is to move away from simple tool application thinking, allowing AI to have a self-evolving system, and letting AI drive AI progress.