Building Smarter Enterprise AI Applications in Java: Agentic Patterns

JavaFest 2026
Udayani V

Udayani V

Senior Member of Technical Staff

Broadcom

Track: Enterprise Java, Spring AI
Session Type: Talk
Edition: JavaFest 2026

Connecting an LLM to Spring Boot is straightforward. But running agents in production quickly hits real-world walls: exploding token bills, context window bloat, task forgetting, and the complexity of coordinating multiple agents across microservices.

In this session, we move beyond basic prompt-and-response interactions to explore production-grade Agentic Design Patterns in Spring AI:

  • Efficiency & Cost Optimization: Slash token overhead using Dynamic Tool Discovery (on-demand tool search) and Prompt Caching.
  • Observability: Eliminate LLM blind guessing with Human-in-the-Loop Clarification (AskUserQuestion) and explainable tool execution with Tool Argument Augmentation.
  • Reasoning & Planning: Solve the “lost-in-the-middle” problem using Self-Governing Planning (TodoWrite) and modular Agent Skills with progressive disclosure.
  • Scale Through Collaboration: Decompose complex workflows into In-Process Hierarchical Subagents and distributed multi-agent systems via the open Agent-to-Agent (A2A) Protocol.

Through live demos and code walkthrough, you will discover how to build modular, maintainable, and intelligent applications using the familiar Spring programming model.