AWS walkthrough: building production-grade autonomous Java agents with Spring AI and Bedrock
AWS ML Blog · rss · 2026-11-12
AWS ML Blog publishes a detailed engineering walkthrough on taking autonomous agents from prototype to production in Java using Spring AI and Amazon Bedrock AgentCore, with two deployable systems:
- Travel assistant (single agent starter): one Spring Boot app exposing REST endpoints and the AgentCore runtime contract (/invocations, /ping), calling Bedrock via the Converse API with @Tool tool calling, SSE streaming, and AgentCore memory for short- and long-term context.
- IDP multi-agent system: a ClassifierAgent routes documents to five specialist agents (Lawyer, Mortgage, Insurance, Tax, General) built on an AbstractDocumentAgent template, extracting structured JSON with confidence scores; a SpEL rules engine dispatches ALERT/BLOCK alerts and routes low-confidence docs to human review.
The post also covers the three agent capabilities (planning, tool use, memory), an 11-stage end-to-end reference architecture (gateway auth, guardrails, routing, human review, observability), and three collaboration patterns: router dispatch, evaluator + rules engine, and parallel fan-out aggregation. Stack: JDK 21, Spring Boot 4.1, Spring AI 2.0, Terraform, ARM64/Graviton images, least-privilege IAM.
More from coding & agent
- Perplexity teammate builds GPU markets terminal using the Computer agent — jeffgrimes9 · 2026-11-12
- AWS open-sources Emulator written 100% by autonomous AI agents — blaizedsouza · 2026-11-12
- Roma 2.0 packs Karpathy's autoresearch into one multi-agent prompt — EXM7777 · 2026-11-12
- From SDLC to ADLC: agents quietly take over planning, coding, testing, and deployment — blaizedsouza · 2026-11-12
- How a $167K-per-deal firm was wired onto one MacBook with Claude in three afternoons — aryanXmahajan · 2026-11-12
- LangChain.js ships support for Gemini 4 Carbon's new Interactions API — m4rkmc · 2026-11-12