AWS guide: Java agents via Spring AI on Amazon Bedrock AgentCore

The walkthrough arrives as Java teams look to apply existing Spring skills to agent development, rather than adopting a separate Python-centric stack. AWS positions Amazon Bedrock AgentCore as framework- and model-agnostic, so teams can bring their preferred agent framework and foundation model while using managed runtime, memory, browser automation, code execution, and identity management. The guide's two examples illustrate a progression: a single travel assistant demonstrates tool calling and conversation memory, while a multi-agent document processing system adds specialist agents, a rules engine, and alerting. Notably, the rules engine lets operators add business checks through configuration rather than code changes, which the guide frames as a safe and auditable extension point. AWS also details resilience patterns such as retries with exponential backoff and circuit breakers, plus security measures like deriving memory actor IDs from verified OAuth 2.0 identities to prevent cross-user access. The guide notes that structured output was verified with Anthropic Claude Sonnet 4.5 on Amazon Bedrock and is best-effort, suggesting teams may need validation and retry logic for stricter schema enforcement.