Resilient AI Agents in Java: A Production Framework for When Reasoning Isn't Enough

AI agents fail in production in ways traditional services don’t: non-deterministic outputs, cascading tool failures, context window exhaustion, and hallucinations that pass unit tests. Most tutorials stop at simple agent execution, but deploying agents into enterprise environments requires a completely different architectural mindset.

This talk presents a battle-tested production framework for building resilient, enterprise-grade AI agents using Java and Spring AI. We cover:

• Circuit breakers & fallback chains for agent reasoning steps
• Context window management and dynamic token budgeting
• Tool call validation and runtime guardrails
• Observability and tracing for multi-step agent decisions using OpenTelemetry and Spring AI observation APIs

Attendees will leave with actionable design patterns, architecture blueprints, and a concrete mental model for taking Java AI agents from prototype to production with confidence.