AWS walks Java teams through Bedrock and Spring AI agents
A new AWS how-to shows Java developers how to build autonomous agents with Spring AI on Amazon Bedrock and deploy them on AgentCore.
What AWS published
On 11 November 2026 the AWS Machine Learning Blog published a technical walkthrough for Java developers by Viquar Khan and Mahammadali Saheb on building autonomous AI agents with Amazon Bedrock and Spring AI. The post argues that Java teams already have JVM tooling, CI/CD pipelines and operational habits in place, but that agent frameworks on their own do not solve scalability, governance, security or operations. It then works through tool calling, conversation memory, resilience patterns and multi-agent orchestration, and deploys the result with Amazon Bedrock AgentCore.
Two reference systems instead of one demo
The post builds two systems that share the same patterns, and it is explicit that one is the entry point and the other is the scaled-up version.
The first is a travel assistant: a single agent with tool calling for weather lookups and itinerary search, streamed responses, conversation memory and resilience handling. The authors recommend it as the starting point.
The second is an intelligent document processing system with five specialist agents named for their domains (Lawyer, Mortgage, Insurance, Tax, General), plus a rules and workflow engine, alerting and observability.
How the single agent is wired
The travel assistant is a Spring Boot application with two entry points: REST endpoints for a React frontend, and the AgentCore runtime contract of POST /invocations and GET /ping. Spring AI's ChatClient orchestrates calls to Amazon Bedrock through the Converse API, and tool invocations are resolved locally in the same process. Tools are @Tool annotated methods, structured output maps to Java records, and Server-Sent Events push tokens to the client as they are generated. AgentCore memory carries both short-term context inside a session and long-term context across sessions, so no external state management is needed. The container image is ARM64, built for AWS Graviton, and infrastructure is provisioned with modular Terraform.
The prerequisites are specific: JDK 21, Spring Boot 4.1 with Spring AI 2.0 and spring-ai-agentcore 2.0, Maven 3.9 or newer, Node.js 18 or newer for the frontend, Terraform 1.5 or newer, Docker with Buildx, AWS CLI v2 and an AWS account.
The definition of an agent the post uses
The article draws a line that is useful when scoping work. A chat completion call produces text but does not decide what to do next, reach external systems or remember earlier sessions. An agent combines planning, tool use and memory, running in a goal-directed loop until the task completes or the agent decides it cannot proceed. A retrieval augmented generation application, by contrast, has retrieval as a narrow form of tool use but no planning loop.
The reference pipeline in the post has 11 stages. A request passes a runtime gateway for authentication and rate limiting, input guardrails screen content, and a router classifies the document and picks a specialist agent. Low-confidence or flagged results go to a human review queue before an answer is returned. Resilience, observability, logging and least-privilege IAM sit across every stage.
Limits and open questions
The post is a how-to, not a benchmark. It gives no latency figures, no cost comparison and no traffic numbers, and it does not say which AWS Regions support AgentCore. Versions for Spring AI and spring-ai-agentcore are listed without release or support status.
What the article infers rather than states is also worth separating. AgentCore is described as framework-agnostic and model-agnostic, with managed runtime, memory, browser automation, code execution and identity management. This suggests the model and framework choices stay swappable behind Spring's usual configuration, but the walkthrough itself only demonstrates the Bedrock path.
What this means for building apps with AI
For teams already on the JVM, the practical message is that agent code does not need a separate stack: it can sit inside ordinary Spring idioms for APIs, security and data access. The two-system structure is the more transferable idea, since it treats a single agent as the testable foundation and multi-agent orchestration as something you graduate to.
Not every AI feature needs that machinery. If you are building a web app and want AI inside it rather than a distributed JVM agent platform, Draftship's guide to adding AI to your app covers the in-app case, where a description in plain language produces an app with a database, sign-in and hosting built in. Draftship is not affiliated with AWS, and nothing here suggests it supports Bedrock, Spring AI or AgentCore.
Primary sources
- Build autonomous AI agents with Amazon Bedrock and Spring AI · AWS Machine Learning Blog, Nov 12, 2026
This article was drafted with AI assistance from the primary sources listed below and checked automatically against them. An editor hasn't reviewed it yet.