Agentic AI applications can be categorized into two types based on their triggering mechanism: user-triggered agents and system-triggered agents. With the rapid evolving of techniques, we believe the future of agentic AI application is industrialized, and the majority of agentic workload would be triggered by real-time system events and data updates.

System-triggered agents must run continuously, act autonomously, and recover from failure without manual intervention. Flink’s capabilities in real-time distributed event processing, state management and exact-once consistent fault tolerance make it well-suited as a framework for building such agents. Furthermore, system-triggered agents are often tightly coupled with data processing. Flink’s outstanding data processing capabilities brings seamless integration between data and agentic processing, differentiate it from other agent frameworks.

Under such a background, the Flink community has initiated a new sub-project, Flink Agents, an Agentic AI framework based on Apache Flink.

This presentation will cover:

• Use cases and requirements of system-triggered agents.
• Why Apache Flink is well-suited for building such
• High-level architectural design of Flink Agents
• Demonstration of building agentic application with Flink Agents

Speakers:

Xintong Song: Apache Flink PMC Member, Staff Software Engineer @ Alibaba Cloud

Xintong is an Apache Flink PMC member. He is also a Staff Software Engineer and leads the Flink Java Engine Team at Alibaba Cloud. Prior to that, he received a Ph.D. degree in computer science from Peking University.