A conference talk by IBM engineers Mario Fusco and Kevin Dubois covering agentic AI patterns implemented in LangChain4J with Quarkus. The talk walks through four stages: building a single AI service, composing agents using deterministic workflow patterns (sequence, loop, parallelization, conditional routing), giving LLMs autonomy via a supervisor pattern, and a new pluggable planner abstraction. Key concepts include the 'agentic scope' (shared state across agents), non-AI agents, human-in-the-loop, goal-oriented planning using dependency graphs, and full composability of all patterns. Code examples demonstrate practical use cases like story generation pipelines, currency exchange, and request routing to specialized expert agents.
•50m watch time
Sort: