Spring AI 2.0 Brings Bedrock AgentCore to Java
Spring AI 2.0 and the Spring AI AgentCore SDK establish a production stack for autonomous agents on Java JDK 21 and Spring Boot 4.x, moving past legacy Bedrock setups.
Spring AI 2.0 and spring-ai-agentcore 2.0 require Java JDK 21 and run with Spring Boot 4.x, modernizing how engineering teams deploy production Java agents. The release integrates the Spring AI AgentCore SDK, an open source library connecting Amazon Bedrock AgentCore infrastructure directly to the Spring AI ecosystem. This pattern shifts development away from Amazon Bedrock Agents Classic, which no longer allows new sign-ups.1234
Anthropic workflows versus Bedrock AgentCore primitives

Anthropic distinguishes between agentic workflows and true autonomous agents. Workflows orchestrate large language models and tools through predefined code paths, whereas autonomous agents dynamically direct their own processes and tool execution. Operating in a goal-directed loop until a task finishes requires 3 core primitives: planning, tool use, and memory, as detailed in our guide on autonomous AI agents runtimes and costs. Tool design also governs reasoning requirements, as general-purpose tools demand more powerful language model reasoning capabilities than specialized tools.156
Implementing the Spring AI 2.0 stack
The Spring AI 2.x.x mainline branch requires Java JDK 21 alongside Spring Boot 4.x. Inside this runtime, Spring AI provides an Advisors API designed to encapsulate recurring Generative AI patterns and transform data moving to and from language models while maintaining functional portability. The Spring AI framework supports major providers, including Anthropic, OpenAI, Amazon Bedrock, Google, Ollama, Mistral AI, and DeepSeek.12
Designing for multi-agent coordination
For complex document workloads, Amazon outlines an Intelligent Document Processing (IDP) architecture running five specialist agents: Lawyer, Mortgage, Insurance, Tax, and General. A ClassifierAgent ingests incoming documents and routes them to the appropriate specialist based on classification rules. To optimize latency across multiple specialists, the system applies a parallel fan-out pattern. This coordination pattern concurrently dispatches one request to multiple specialist agents and aggregates their outputs before triggering a downstream rules engine.1
Operationalizing the AgentCore runtime contract
Amazon Bedrock AgentCore provides a model-agnostic and framework-agnostic platform built to operate agents at scale. Teams evaluating infrastructure alongside self-hosted AI agents must satisfy the AgentCore runtime hosting contract by having their Spring Boot application expose two specific endpoints: /invocations for task routing and /ping for operational health checks. For fleet management, the AgentCore infrastructure supplies native monitoring and observability in a standardized OpenTelemetry compatible format.17