MuleRun CTO Junliang Shu: Model Gap Rapidly Converging, Agent Moat Shifting to "Speed + Data"

April 21st – During the “Decoding Web 4.0: When AI Agents Take Over On-Chain Authority” roundtable, MuleRun CTO Steven Hu dove into the “Agent Moat” topic, noting the traditional idea of an AI tech moat is quickly fading. The core driver? Faster convergence of model capabilities and exponential gains in development efficiency. Hu pointed out the performance gap between mainstream large models is narrowing rapidly—especially over the past year, when the capability gap between domestic and global models has shrunk significantly. Meanwhile, exploding coding capabilities have supercharged software development: features that once took weeks or months to build now take days. This means Agent frameworks and specific functional modules can be quickly reused or replicated via open-source tools, making product-level “feature moats” increasingly fragile. In this landscape, Hu argues Agents’ future core competitiveness will hinge on two pillars: 1. **Continuous high-frequency iteration**: Can teams maintain a leading product update pace long-term? 2. **Data advantages**: Exclusive data resources (e.g., industry-specific/regional data) and user data accumulation. Long-term user behavioral and memory data (hard to migrate) boosts stickiness and competitiveness. He concluded that as models and technologies gradually “level out,” the Agent moat is shifting from “technical capability” to a comprehensive battle of “data assets and execution efficiency.”