From native Java APIs to AI-assisted development, this week examines how engineering platforms are evolving.
🎯 Featured Article
🧩 JEP 540 Proposed to Target JDK 28 with a Simple JSON API A compact, JDK-provided JSON API for common parsing, navigation and generation tasks without external dependencies.
- The API targets RFC 8259 JSON without data binding or streaming.
- The incubating API could still change or be removed.
⏩ TL;DR (Quick Recap)
- Java may gain a focused native JSON API in JDK 28.
- JVM libraries continue evolving around compatibility and performance.
- Spring AI simplifies agentic workflows and retrieval techniques.
- AI tooling raises new questions around cost, infrastructure and code quality.
☕ JVM Corner
🛠️ Evolving a Java MCP Server During MCP Specification Upgrades
A compatibility-first approach helps Java MCP servers adopt specification changes without disrupting existing users.
- New MCP versions refine discovery, caching and typed outputs.
- Compatibility helps servers evolve alongside the protocol.
🧭 Signatures, be true: domain errors and functional handling in Kotlin
Kotlin sealed errors and Either make expected domain failures explicit in function signatures and easier to handle.
- Sealed types make domain errors visible to callers.
- Either supports explicit functional error handling.
⚡ When @Immutable Made All the Difference: A 3.2x Speedup with Hibernate Second-Level Cache Hibernate second-level caching delivered another major performance improvement, when combined with immutability and cascade configuration.
- Three annotations were required for the full cache benefit.
- Partial configuration left performance more than eight times slower.
🔐 Preparing for Change: Safe Switching over Sealed APIs OpenJDK’s exhaustiveness guidance explains how sealed APIs should account for future subtype additions and runtime MatchException.
- Exhaustive switches protect against currently known subtypes.
- API evolution should guide clients on future subtype handling.
🍃 Spring Updates
🤖 Spring AI Recipe: Reasoning and Acting Spring AI’s ChatClient and tools can implement the ReAct reasoning-and-acting pattern without requiring a dedicated agent framework.
- Tool use provides a foundation for agentic workflows.
- ReAct combines reasoning with actions against available tools.
🔎 Spring AI Recipe: Better RAG Results with HyDE HyDE improves retrieval by generating a hypothetical answer first and using it to find more relevant documents.
- Query wording can prevent standard RAG from finding useful content.
- Hypothetical answers provide an alternative retrieval representation.
⚙️ Spring Boot Configuration Management Best Practices Spring Boot’s externalized configuration lets one application artifact operate across environments using multiple configuration sources.
- Environment variables and property files support deployment flexibility.
- Command-line and system properties provide additional configuration sources.
🔍 Extra Reads
Build and Delivery
📦 Solving Gradle metadata and Renovate integration Gradle metadata details can affect how Renovate discovers and updates project dependencies.
🧪 Lessons Learned from Fixing Flaky Tests with Claude Claude helped eliminate most flaky integration tests after parallel execution delivered a tenfold speedup.
AI Economics and Infrastructure
💸 Claude Code pricing: same tokens, same model, up to 40x the price
AI coding agents can consume dramatically more tokens, making usage economics a significant engineering concern.
🏢 No, local models will not win AI datacenters will remain the dominant destination for inference despite advances in local models.
🔇 AI;DR (AI; Didn’t Read) AI;DR proposes a way to filter out low-value AI-generated content and avoid increasingly large volumes of automated text.🧠 Qwen 3.8 27B is excellent, but it defaults to wildly overthinking things
Qwen 3.8 27B offers a capable Apache 2.0-licensed model at a size practical for reasonably equipped laptops.
Developer Platforms and AI Tooling
🐹 Go 1.27 is released Go 1.27 introduces updates across the language, compiler and toolchain, runtime and standard library.
🏗️ Why Go is an Ideal Language for AI-Assisted Software Engineering
Go’s simplicity and clear boundaries can help engineers review and maintain code increasingly generated by AI assistants.
🔀 OpenRouter is Joining Stripe OpenRouter is joining Stripe, while continuing its model marketplace and gateway for accessing hundreds of AI models.
🗄️ Origin Code Hosting Cursor’s Origin adds agent-oriented code hosting with repositories, pull requests, browsing and GitHub synchronization.
Originally posted on marconak-matej.medium.com.