Java meets Artificial Intelligence

Latest Java AI News

Latest news and releases from across the Java AI ecosystem — new framework versions, MCP updates, and inference engines.

• LangChain4j 1.22.0 Released — adds OpenAI Decisions API support for decision models (OpenAiDecisionModel), BedrockBatchChatModel for AWS Bedrock batch inference, and automatic prompt caching with cache TTL for the Anthropic integration
• LangChain4j 1.21.0 Released — introduces decision models, a new type that answers questions with typed responses and probabilities instead of generated text, for classification and routing
• LangChain4j 1.20.2 Released — patch release with A2A fixes
• One Database, Two Models — A N M Bazlur Rahman explores MySQL JSON Duality Views in Java: simpler document reads/writes over JDBC, optional-etag conflict handling, comparisons with JPA/Spring Data JDBC, and read-only LangChain4j tools backed by local Ollama for AI agents
• Quarkus Flow 1.1.3 Released — patch release fixing while-mode loops in LoopAgenticFlow, removing the in-memory/MVStore lifecycle-management module, and dropping the unused Dev UI extension descriptor
• BoxLang AI 3.4.0: Gateways, Human-in-the-Loop, and a Full Security Stack — Luis Majano details platform gateways, a restructured Human-in-the-Loop subsystem with durable decisions and batched approvals, layered prompt-injection/data-loss security controls, and expanded AWS Bedrock support
• Spring AI 2.1.0-M1 Available Now — adds a Message Parts content model (ordered TextPart/ReasoningPart/ToolCallPart) for round-tripping reasoning and tool calls, a new OpenAiResponsesChatModel for OpenAI's Responses API, and VectorStore.upsert() support for pre-computed embeddings; baseline moves to Spring Boot 4.2
• Quarkus LangChain4j 1.14.0 Released — adds a new @McpClientAgent annotation, fixes MCP SSE parsing and A2A agent discovery via Apicurio Registry, and upgrades to LangChain4j 1.20.1
• LangChain4j 1.20.1 Released — methods annotated @McpClientSupplier can now also accept supplier parameters
• Koog 1.3.0 Released — adds ready-made LLModel definitions for the latest Anthropic, OpenAI, Google, Bedrock, DashScope, DeepSeek, Mistral, Ollama, and OpenRouter models, and fixes Google prompt-cache token accounting and Langfuse/Gemini trace display
• TypeSafe Jev Meets Apache Camel — walks through six practical examples of wiring TypeSafe's Jev semantic decision model into Camel routes for fast, structured classifications and judgments in integration workflows
• TornadoVM 7.0.0 Released — adds a TileContext API for Java CUDA tile-based GPU compute, a cuDF library-task provider, matrix multiply-accumulate improvements, and fixes half-precision (FP16) conversion and storage across the CUDA, OpenCL, and Metal backends
• B.Y.O.A. — Camel 4.23 adds support for bringing your own coding agent, via the Agent Client Protocol, into Camel TUI, paired with Camel Kit to build and inspect integrations in one conversation
• Camel's Local-Model Coding Benchmark, Round 2 — follow-up to the original local-model benchmark post: 31 of 33 found issues fixed in Camel 4.23, with step-by-step accuracy improving from 81% to 92%
• Spring AI and TypeSafe Jev: Fast, Cheap, Structured Decisions — introduces Spring AI TypeSafe 0.1.0, a Spring AI Community integration with TypeSafe AI's hosted Jev API for typed classification, scoring, judging, guardrails, RAG filtering and reranking, and tool selection without generating text
• MCP Goes Stateless, and Quarkus Already Implements It — Kevin Dubois explains how the MCP 2026-07-28 spec revision removes required session state so any server instance can handle any request, and how the Quarkus MCP Server's 2.0.x line already supports it while staying backward-compatible with stateful clients
• Google ADK for Java 1.10.1 Released — adds an @Experimental annotation for unstable APIs, session-resumability checkpoints, and a one-way ADK Java-to-Kotlin interop module, plus fixes for dev UI routing, memory-search timestamps, and Vertex AI session handling
• Quarkus Flow 1.1.2 Released — patch release adding missing native-image classes, fixing OIDC-blocking and topology issues, and adding missing ServiceLoader registration for URITemplate
• Authorizing What an AI Agent May Do in Apache Camel — rather than relying on prompts for security, enforces AI agent tool calls at the boundary using SPIFFE for identity and Open Policy Agent (OPA) for authorization, so a camel-ai-tool-exposed route can safely expose different capabilities to different callers based on verified identity
• What's New in the September 2026 Azul Payara Release? — MicroProfile 7.1 lands across Payara's product lines, Jakarta Agentic AI support expands to Payara Micro and Embedded (an agent as a CDI bean annotated with @Agent), and a jackson-databind security patch ships across all supported branches
• Embabel Agent 1.5.3 Released — adds role-based default LLM configuration, strict-mode native structured output, JDBC Blackboard persistence configuration, retrying of failed embedding batches, and moves the MCP server from SSE to Streamable HTTP
• Embabel Agent 1.5.2 Released — adds agent process persistence with snapshot/restore functionality backed by JCache storage, improvements to LLM retry policies and structured output APIs, and a new Podman skill script execution engine
• How I Built an AI Assistant for My Career with Java, Spring AI, and MongoDB — Ricardo Mello builds "RicAI," a Spring Boot career assistant combining Spring AI's VectorStore (MongoDB Atlas + OpenAI embeddings) for semantic search, a ChatClient with MongoDB-backed conversation memory, and @Tool-annotated methods for structured queries and Calendly meeting scheduling
• Can You Explain the Diff Your AI Agent Just Wrote? — Viktoria Evdokimova argues passing tests prove correctness but not comprehension, and shows how JetBrains-IDE agents like Explyt surface debugger state and structured test results — with a rule of thumb: if you can't explain what an AI-generated diff breaks when reverted, you don't understand it well enough to merge it
• We Had a Frontier AI Coach a Small Local Model Through Camel — a 22GB local model on a laptop was run twenty times over two days against Camel's beginner examples, with fixes applied to Camel between every run based on what made it fail, shipping in Camel 4.23 as better Simple-parser/YAML-loader error messages, a validator that says what to write, and catalog tools that answer the question asked
• Quarkus LangChain4j 1.13.3 Released — upgrades to LangChain4j 1.19.3, fixes a module-name duplicity issue in the Chat Scopes WebSocket component, and adds automatic AgenticScope-serialization registration for domain objects referenced in agentic method signatures
• Quarkus Insights #259: MCP Servers Without the Boilerplate — Introducing reShapr — Laurent Broudoux (Microcks, Postman Open Innovation & AI) demos reShapr, a declarative platform that turns existing OpenAPI/GraphQL/gRPC APIs into production-grade MCP servers by configuration instead of code generation, citing 97% fewer tokens consumed in real-world workflows
• InfoQ Java News Roundup — weekly Java ecosystem roundup covering the Jakarta Agentic AI 1.0 specification's M1 milestone release and new project website, Google's ADK for Kotlin reaching 1.0 GA, Jakarta CDI 5.0 GA, and point releases across TornadoVM, RefactorFirst, Groovy, and Gradle
• A2A Jakarta 1.0.0.Final Released — first stable release of Jakarta integrations for the A2A Java SDK's JSON-RPC, gRPC, and HTTP+JSON/REST transports, built on a2a-java 1.3.2.Final, with a new example demonstrating A2A agent deployment across multiple WildFly cluster nodes using JPA and Kafka for shared state
• TornadoVM 6.1.0 Released — adds cuRAND library support for device-side random number generation and CUDA backend performance optimizations, including skipping the per-launch kernel stack-frame upload when the device copy is unchanged
• ADK for Kotlin 1.0 Announced — Google's Agent Development Kit for Kotlin reaches full feature parity with ADK Core, adding on-device inference via LiteRT-LM and ML Kit, Firebase AI Logic for hybrid cloud workflows, Room/AppSearch-backed state persistence, and Kotlin Symbol Processing for compile-time tool-schema generation
• One Base URL, Many Coding Agents — Yurii Kostyukov shows how to probe an OpenAI-compatible AI gateway's availability from Java's java.net.http client, plus token-estimate calibration and credential-refresh lessons from building a multi-agent IDE gateway
• A2A Java SDK 1.3.2.Final Released — patch release fixing the HTTP client's PostBuilder to not follow redirects by default, plus logger field renames and test cleanup
• "Commit Created" — But It Isn't — Viktoria Evdokimova analyzes public issue trackers from Claude Code, Copilot, and other coding agents to identify three root causes of agents falsely reporting completed work, and gives under-a-minute independent verification checks (git log, ls -la, run panels) developers can run against agent claims
• Debugging Is Invariant Discovery — a Kotlin/Spring Boot/Kafka case study showing an AI agent was tactically useful (queries, log reads) but strategically passive without runtime access, and how making implicit business invariants explicit as tests helps both humans and agents spot related failures
• AgentScope Java 2.0.3 Released — adds Anthropic prompt caching, a deliver_artifact tool for sandboxed agents, FinalAnswerFilterMiddleware for ReAct streams, state versioning for concurrency control, an AgentProtocolEventBus for SSE handling, and reliability fixes across the reasoning loop, harness, and sandbox layers
• InfoQ Java News Roundup — weekly Java ecosystem roundup covering TornadoVM 6.0 with Foreign Function & Memory API support, LangChain4j 1.20.0's reactive AI service methods, and point releases across JReleaser, Java Operator SDK, JHipster, and Yupiik Fusion
• LangChain4j 1.20.0 Released — adds non-blocking/reactive support for AI Services (methods can return CompletableFuture or Flow.Publisher), opt-in Jackson 3 support, and fixes across MCP, embedding stores, and Azure integrations
• From Spec-Driven Development to Living Specifications in Java Projects — Maximillian Arruda introduces SDD4J, a workflow keeping specifications synchronized with code via architecture adapters, testable EARS requirements, and specification-to-test traceability, without forcing a single architectural style
• Did Your AI Agent Run the Debugger? — a case study comparing two Claude agent runs fixing the same JVM bug in a JetBrains plugin, showing that reproducing a failure at a breakpoint beats reasoning from stack traces alone — fewer tokens, no unnecessary API changes
• Building an Agentic Warehouse Management System — Part 3: Tools, Decisions, and Actions — Ricardo Mello shows the agent accessing operational data, retrieving business policies via semantic search, deciding on replenishment, and acting through controlled tools that create requests and draft notifications within application-defined boundaries
• Building an Agentic Warehouse Management System — Part 2: Java and Spring AI — implements a warehouse-replenishment agent with Spring AI using a planner/executor/reporter architecture, where the planner turns high-level goals into structured, MongoDB-stored execution plans built from the agent's declared capabilities rather than hardcoded logic
• Building an Agentic Warehouse Management System — Part 1: Where AI Agents Add Value — Ricardo Mello opens the series by examining inventory-replenishment decisions that need contextual analysis beyond fixed rules, and how an agent orchestrates tools within defined boundaries rather than replacing deterministic business logic
• Google Gen AI SDK for Kotlin 1.0 Announced — a Kotlin Multiplatform (JVM, Android) SDK for the Gemini Developer API and Gemini Enterprise Agent Platform, with coroutines-based streaming, multi-turn chat, automatic function calling, and the Live API, published to Maven Central
• A2A Java SDK 1.3.1.Final Released — patch release fixing a missing beans.xml in the multitenancy module, makes @PublicAgentCard optional with a fallback, and adds 404 handling for unknown tenants
• AgentScope Java 2.0.2 Released — routes tasks through AgentFactory, adds RuntimeContext force-sync for agent spawning, parent session ID tracking and full event forwarding for remote subagents, and decouples TaskStore from the execution WorkspaceManager
• TornadoVM 6.0.0 Released — major release adding a CUDA C backend via NVRTC, native FP8/BF16 tensor support, and a hybrid API with cuBLAS/cuDNN/cuFFT library-task providers, while removing the legacy PTX, SPIR-V, and FPGA backends
• BoxLang AI 3.4.0 Released — adds gateway session management for wiring agents to inbound channels, live run control (cancel/steer in-flight runs), batch tool-call approvals, prompt-injection/PII guard middleware, and AWS Bedrock provider parity fixes
• Observe Your Camel AI Routes with GenAI OpenTelemetry — introduces Camel's new GenAI observability feature, emitting OpenTelemetry spans and Micrometer metrics for LLM routes, demoed via the Camel CLI/JBang with Ollama
• GenAI Observability with Spring Boot and the Camel Observability Stack — productionizes Camel's GenAI observability with Spring Boot, Prometheus, VictoriaTraces, and Perses
• Quarkus LangChain4j 1.13.1 Released — bugfix release fixing the agentic Dev UI Topology/Executions pages, preserving generic tool argument types, Infinispan metadata filter handling, and streaming fixes for JAX-RS and MCP HTTP transports
• Koog 1.2.0 Released — adds a skills module implementing the Agent Skills specification (repository discovery, catalog prompt generation in XML/YAML/JSON) and a koog-bedrock-agentcore-runtime module for running Koog agents as Amazon Bedrock AgentCore Runtime entry points, plus OpenAI Responses API and reasoning-content fixes
• A2A Java SDK 1.3.0.Final Released — adds multitenancy support with per-tenant routing and a functional-interface AgentExecutor with a default cancel method, alongside security hardening: CR/LF header-injection prevention, SSRF protection for push-notification URLs, and fail-closed task-authorization enforcement
• Quarkus Flow 1.0.0 Released — first stable release of Quarkus's embedded workflow engine, defining workflows in YAML (Open Workflow Specification) or a Java DSL, with LangChain4j integration for agentic pipelines (branching, retries, human-in-the-loop), CloudEvents/Reactive Messaging support, and GraalVM native compilation
• Azul Payara Community 7.2026.8 Previews Jakarta Agentic AI — first runtime preview of the Jakarta Agentic AI 1.0 API, letting developers build AI agents on Jakarta EE using standard CDI beans and dependency injection
• Embabel Agent 1.5.1 Released — RAG improvements including a SectionReader for document navigation, embedding-driven skills, Atlas Cloud BYOK support, thinking-tag selection control, and fixes for Bedrock chat options and Ollama thinking-mode extraction
• Micronaut Framework 5.1.2 Released — maintenance release with updates across Core, Cache, Test, Data, Security, Serialization, and Flyway modules
• Spring AI 2.0.1 Available Now — first maintenance release since the June 2.0.0 GA, fixing 80+ issues including 7 CVEs (PDF processing, cache, session allocation, file operations, semantic caching, Redis, tool dispatch), adding configurable tool-call limits and OpenAI audio streaming, and removing deprecated Mistral AI chat models
• MCP Java SDK 2.0.1 Released — first patch on the 2.0 line: upgrades the conformance suite to 0.1.16 and Spring AI conformance to 2.0.0 GA, bounds HTTP and STDIO server/client reads, and fixes SSE event classification, pagination, and unregistered-handler behavior in stateless servers
• A2A Java SDK 1.2.0.Final Released — hardens the authorization model with read-authorization enforcement on referenced task lookups, adds a TaskStreamLifecycleHook SPI for stream lifecycle management, and ships breaking changes including method renames and a TaskState enum change
• Quarkus LangChain4j 1.13.0 Released — upgrades to LangChain4j 1.19.0 and the latest MCP client spec, replaces the Qdrant gRPC client with a REST client, adds GPULlama3 tool-calling support, and expands the Dev UI with a Guardrails page and RAG context visibility
• DevOps Patterns and Java 26 for On-Premises LLM Platforms in Safety-Critical Environments — Cornelius May covers RAG-on-Kubernetes architecture, Java 26's AOT caching, Vector API, and Structured Concurrency, and treating prompts as versioned, governed code
• Vibe Coding, Maven, and the Dependencies You Didn't Choose — Steve Poole warns that AI coding assistants are silently choosing an app's Maven dependencies, transitive packages, and plugins, and argues supply-chain scrutiny needs to keep pace with AI-accelerated dependency selection
• Micronaut Framework 5.1.1 Released — maintenance release covering Micronaut Core and Micronaut OpenAPI
• Spring AI Recipe: Reasoning and Acting (ReAct) — Craig Walls shows how Spring AI's ChatClient with tools implements the ReAct pattern via an automatic ToolCallingAdvisor loop — the LLM reasons, invokes a tool, observes the result, and iterates — demonstrated with a small Disney Parks example
• Building AI-Driven Java Systems with Spring AI and Java 26 — treats LLMs as first-class components with explicit responsibilities, leveraging Java 26's Virtual Threads and Structured Concurrency for efficient I/O-bound agentic workflows
• LangChain4j 1.19.0 Released — adds MCP client improvements, Anthropic Batch Chat Model support, a Milvus V2 service update with hybrid search, watsonx.ai Model Gateway support, and Google GenAI thinking support
• Spring AI Recipe: Building a Graph-Based Agentic Workflow with LangGraph4j — Craig Walls builds a customer-support workflow with LangGraph4j nodes for classification, billing, and technical support integrated with Spring AI, contrasting LangGraph4j's framework-neutral graphs with Spring AI Alibaba Graph
• Evolving a Java MCP Server During MCP Specification Upgrades — Ana-Maria Mihalceanu shows how to adopt the MCP 2026-07-28 spec while staying compatible with older clients, routing new stateless requests through an adapter layer alongside the legacy initialized workflow
• Embabel Agent 1.5.0 Released — migrates to Spring AI 2.0 GA (Spring Boot 4, Jackson 3), adds Alibaba Cloud DashScope model support, RAG chunk-metadata and full-text scoring improvements, and a vendor-neutral streaming tool loop
• Spring AI Recipe: Filtering RAG Results with Metadata — Craig Walls shows how attaching metadata to document chunks and filtering vector-store queries with Spring AI's VectorStore and QuestionAnswerAdvisor narrows RAG retrieval before the LLM sees it
• Introduction to Retrieval-Augmented Generation with Java and MongoDB — tutorial building a JAX-RS/Helidon RAG service with LangChain4j and MongoDB Atlas as a combined vector store and operational database, retrieving context to answer HR policy questions before calling the LLM
• A2A Java SDK 1.2.0.Final Released — adds programmatic auth wiring and a TaskStreamLifecycleHook, upgrades to Quarkus 3.37.3, and ships breaking changes: cross-module split-package resolution, TaskState.UNRECOGNIZED renamed to TASK_STATE_UNSPECIFIED per spec, and corrected authorization enforcement on referenced task lookups
• Camel Routes as AI Tools: Unified Tooling and MCP Server in Camel 4.22 — introduces camel-ai-tool, a unified tool abstraction so a route defined once works with LangChain4j, Spring AI, or OpenAI, and camel-mcp-server, which exposes tagged routes directly as MCP tools
• AgentScope Java 2.0.1 Released — maintenance release expanding model-provider support (DeepSeek, GLM, Kimi, MiniMax) and hardening Harness subagent/permission handling, plus fixes for memory leaks and streaming tool-arg repair
• InfoQ Java News Roundup — weekly Java ecosystem roundup covering two new JDK 28 JEPs, Jakarta Agentic AI's first milestone release, GPULlama3.java 1.0.0 GA, GraalVM 25.2's new Graal Script Agent, and point releases across Micronaut, Quarkus, and JobRunr
• Jakarta Agentic AI 1.0.0-M1 — first milestone release of the new Jakarta EE agentic AI specification, adding TCK behavioral-test infrastructure and module/javadoc improvements ahead of 1.0
• Jakarta Agentic AI Hits Its First Milestone — Dominika Tasarz walks through the 1.0.0-M1 release now on Maven Central: the draft spec's annotation model (@Agent, @Trigger, @Decision, @Action, @Outcome, @HandleException), a new CDI @WorkflowScoped context, and a vendor-neutral LargeLanguageModel facade
• How to Create a Spring Boot Fraud Scoring Service — builds a production credit-card fraud detection REST API with Spring Boot and the pure-Java Deep Netts deep-learning library, embedding the model directly in the app instead of a separate Python inference server
• Under the HAT: Empowering GPU Acceleration for Java — Juan Fumero introduces the Heterogeneous Accelerator Toolkit, which uses OpenJDK Project Babylon's code-reflection APIs to translate sections of Java programs into CUDA and OpenCL at runtime, managing generated code and data migration automatically
• Connecting Java Reinforcement Learning to Python Gymnasium — bridges Java RL code to Python's Gymnasium library via JavaCPP, covering Python lifecycle management and NumPy data transfer, starting with CartPole before scaling up to CARLA
• GPULlama3.java 1.0.0 Released — first major release of the GPU-accelerated Java LLM inference engine, now on Maven Central as io.github.beehive-lab:gpu-llama3 with auto-activating 1.0.0-jdk21 and 1.0.0-jdk25 builds, making TornadoVM-backed GPU inference a plain Maven or Gradle dependency
• Micronaut LangChain4j 2.2.0 Released — adds LangChain4j agentic support, Chroma embedding-store support, guardrails resolved from Micronaut beans, and evaluation testing to the official Micronaut integration
• JetBrains Air Adds Java and Kotlin Code Intelligence — the agent-first tool gains language intelligence powered by the IntelliJ IDEA code engine (Beta) for mixed Java/Kotlin projects — jump-to-definition, find usages, symbol search, and error highlighting on agent-modified code — plus local models via Ollama and LM Studio through ACP-compatible agents
• Spring AI AgentCore 2.1.0 Released — latest release of the Spring Boot integrations for Amazon Bedrock AgentCore, following the 2.0.0 GA
• Micronaut Framework 5.1.0 Released — Micronaut LangChain4j upgraded to 2.2.0 with Oracle chat-memory auto-configuration, Chroma embedding-store support, bean-resolved AI-service guardrails, and agentic support, alongside a Micronaut MCP update to track the latest MCP Java SDK
• When Prompts Become Plugins: Generating User Extensions with Graal Script Agent — introduces a library for turning end-user prompts into sandboxed, reusable JavaScript or Python extensions that run locally within application-defined boundaries
• Pairing In-Process and Hosted Embeddings for Java MCP Tool Development — designing an MCP server on Helidon that supports both local (DJL/MiniLM) and hosted (OpenAI) embedding models behind a stable tool interface
• Embabel 1.0.0 Reaches GA — the JVM agent framework's first stable release promotes experimental RAG APIs to production status, adds generic media/document support, exposes MCP server health via Spring Boot Actuator, and now resolves entirely from Maven Central
• TornadoVM 5.2.0 Adds Native FP8 Tensor-Core Support — packed half2 support and native FP8 conversion with FP8/BF16 tensor-core MMA for the CUDA backend, plus expanded BFloat16 array support and new activation-function epilogues (SiLU, Sigmoid, Tanh, HardSwish)
• Inside Java Podcast: AI Solutions with Spring AI 2.0 — Dan Vega and Lize Raes discuss Spring AI 2.0, deterministic agents, and MCP servers/skills, recorded at JavaOne 2026
• AI Found the Bugs. Who's Patching Your EOL Java Code? — argues AI-assisted vulnerability discovery is now outpacing human-speed patching for unsupported Java dependencies, raising security-economics questions for the ecosystem
• Koog 1.1.1 Released — adds Spring Boot 4 / Spring AI 2.0 starters and a WebClient-based HTTP client, non-text tool results (images/files), Amazon Bedrock memory auto-discovery, and a fix for a Spring AI 2.0 ClassCastException on generic ChatOptions
• Solving Spring AI's UI Challenge with AG-UI's Java SDK — walks through bridging Spring AI agent backends with frontend UIs like CopilotKit via a standardized streaming protocol
• LangChain4j 1.18.0 Released — introduces a Belief-Desire-Intention (BDI) agentic pattern, crash-resilient Human-in-the-Loop suspend/resume for agentic systems, OpenAI TTS support, and a revamped embedding-model API with per-call parameters and multimodal input
• JVM Pulse: Profiling Java with GitHub Copilot — new open-source Copilot canvas extension from Bruno Borges automates GC/JFR profiling and hands off findings to Copilot for AI-driven tuning recommendations
• Compiling OPA Policies to WebAssembly for Quarkus LangChain4j Guardrails — sub-millisecond prompt-injection pattern matching via Open Policy Agent rules compiled to Wasm, falling back to an LLM only for inconclusive cases
• Spring AI AgentCore SDK v2.0.0 Released — major version release of the open-source Spring Boot integration for Amazon Bedrock AgentCore, with auto-configured invocation endpoints, SSE streaming, and short/long-term memory support
• AgentScope Java v2.0.0 Tagged on GitHub — formal GA release tag for the dual-layer ReActAgent/HarnessAgent architecture, a unified 28-event observability model, and a new permission engine
• LangChain4j Agentic Workflows: From AI Calls to Multi-Agent Systems — tutorial covering sequential, loop, parallel, and conditional workflow patterns, goal-oriented planning, and supervisor-based multi-agent systems in Java
• Building AI Systems with MongoDB: Implementing the Planning Pattern — tutorial building a travel-itinerary AI agent with Jakarta EE, Jakarta Data, LangChain4j, and MongoDB using a Reason-Act-Observe planning loop
• A2A Java SDK 1.1.0.Final Released — first feature release after GA, adding a TaskAuthorizationProvider SPI for per-user task authorization in multi-tenant deployments, plus a new project website