Li Zhou
Senior Technical Expert at Alibaba Cloud
With over 10 years of R\&D experience in the middleware domain, he is a core member of the founding team of Apache RocketMQ. He has long led the architecture design, performance optimization, and stability assurance of ultra-large-scale distributed messaging systems, and has been deeply involved in critical technical challenges for major events such as the Double 11 Shopping Festival. He has extensive hands-on experience in building high-concurrency, high-availability systems. Currently, he is dedicated to technological innovation and implementation of AI-Native messaging middleware, focusing on new messaging architectures for the era of large models. His work centers on advancing technologies for model inference scheduling, asynchronous communication between agents, and intelligent task flow orchestration, driving the continuous evolution of messaging middleware toward greater intelligence.
Topic
Apache RocketMQ × AI: Event-Driven Architecture for Asynchronous Agents
The characteristics of AI applications, such as high latency, long execution cycles, strong statefulness, and resource intensity, pose unprecedented engineering challenges to the underlying infrastructure. Traditional message queues are showing limitations in coping with the dynamic, stateful, and resource-sensitive nature of AI-native scenarios. To address this transformation, Apache RocketMQ has initiated a strategic upgrade, evolving from general-purpose messaging middleware into a “message engine” for the AI era. This talk will introduce how to leverage the new features of Apache RocketMQ to build asynchronous Multi-Agent systems. We will explore decoupled communication, context isolation, state recovery, and task orchestration mechanisms between agents, and present practical cases demonstrating how RocketMQ enables task scheduling for Multi-Agent systems. The outline includes: 1) mainstream Multi-Agent development paradigms, including fundamental concepts and common design approaches; 2) solutions for building asynchronous Multi-Agent systems and key challenges of current asynchronous approaches, such as conversation isolation, context management, resource scheduling granularity, and retry strategies; 3) RocketMQ innovations for Multi-Agent scenarios, including million-level queues, lightweight resource management, and event-driven prioritization; 4) how to build a Multi-Agent system based on the new features of RocketMQ.
