MuleRun CTO Shu Junliang: Model gaps converge quickly, and the agent moat shifts to "speed + data"

BlockBeats News, April 21 — During the roundtable discussion “Decoding Web 4.0: When AI Agents Take Over On-Chain Permissions,” MuleRun CTO Shu Junliang stated that the traditional AI technology moat is rapidly weakening, primarily due to the accelerated convergence of model capabilities and the exponential improvement in development efficiency.

He pointed out that the performance gap between mainstream large models is shrinking quickly, especially over the past year, with the capability gap between domestic and international models significantly converging. At the same time, with the explosive growth in coding ability, software development efficiency has greatly increased — functions that previously took weeks or even months to complete can now be achieved within days. This means that both agent frameworks and specific functional modules can be quickly reused or replicated through open-source solutions, making the product-level “feature moat” increasingly fragile.

Against this backdrop, Shu Junliang believes that the future core competitiveness of agents will mainly be reflected in two aspects: first, the ability to iterate at a high frequency continuously, i.e., whether teams can maintain a leading product update speed over the long term; second, data advantages, including exclusive data resources and user-accumulated data. On one hand, platforms with unique data acquisition capabilities (such as data from specific industries or regions) will form natural barriers; on the other hand, the behavioral and memory data accumulated from users’ long-term use on the platform will become key assets that are difficult to transfer, further enhancing user stickiness and product competitiveness.

He summarized that, as models and technologies gradually become “equalized,” the moat of agents is shifting from “technical capability” to a comprehensive competition of “data assets and execution efficiency.”