Agentic Systems: How Liferay AI Hub Coordinates Many Agents
Presenter: Feliphe Costa Marinho, Sr Software Engineer, Liferay
Coordinating many agents in a real product raises hard questions: who decides which agent runs, how do they share context, and how do you keep the whole thing from spinning out of control?This talk uses Liferay AI Hub as a working case study in answering those questions. Building on langchain4j’s agentic framework, AI Hub distinguishes — as the framework does — between deterministic workflows (sequential, parallel, conditional, loop) and pure agentic orchestration, where an LLM-driven supervisor analyzes a request and dynamically plans a sequence of agent invocations. We’ll walk through how AI Hub puts a supervisor agent in front of a fleet of specialized sub-agents, routing each request to the best-matching agent, sharing conversation state through a summarized chat memory, and bounding runaway plans with a max-invocation depth.We’ll then go a layer deeper into what makes this more than a demo: each agent is backed by a Kaleo workflow, so a single agent can chain LLM steps and AI-decision nodes, all running async and streaming results back to the user in real time.Attendees will leave with concrete answers to the questions the talk opens with — who decides which agent runs, how agents share context, and how to keep coordination under control — backed by a real implementation.