Better developer tooling, safer AI-assisted coding and approaches to building software.
🎯 Featured Article
🧠 Java and Kotlin by IntelliJ IDEA: The LSP-Based Extension Reaches Release Candidate Java and Kotlin intelligence is moving beyond the IDE, bringing project-aware code understanding to editors such as VS Code and Cursor.
- The Language Server Protocol extends Java and Kotlin intelligence to third-party editors.
- Project-aware code understanding supports developers and coding agents across different workflows.
⏩ TL;DR (Quick Recap)
- IntelliJ IDEA expands Java and Kotlin intelligence beyond its own IDE, while Java desktop development retains a modern foundation.
- Teams need explicit task requirements and review practices to ensure coding agents deliver the intended changes.
- New AI-assisted vulnerability discovery and database locking considerations highlight the importance of correctness.
- Local models, structured judgment, code execution, and hard spending limits offer different approaches to controlling cost and complexity.
☕ JVM Corner
⚡ Mastering SIMD with Java Vector API Java’s incubating Vector API enables data-parallel computation, offering a way to improve the performance.
- The Vector API is available through the jdk.incubator.vector module.
- SIMD optimization can improve throughput by processing multiple data elements in parallel.
🖥️ The state of java desktop Desktop development combines mature libraries and cross-platform compatibility with packaging tools and local-first application design.
- jpackage simplifies packaging Java applications for distribution.
- Established libraries, desktop compatibility and tools such as JBang support desktop development.
🤝 Green Build, Wrong Task: A Working Agreement for Java Teams Using AI Coding Agents AI coding agents can produce passing tests while misunderstanding the requested outcome, making precise task definitions and working agreements essential.
- Passing builds do not guarantee that an implementation satisfies the original requirement.
- Explicit acceptance criteria help align agent-generated tests with the intended behavior.
🛡️ IBM & Red Hat Find More Than 400 New Vulnerabilities In Popular Java Code Red Hat’s Lightwell initiative uses specialized AI agents to accelerate vulnerability discovery.
📰 Java News Roundup: JobRunr 9, OpenXava 8, Quarkus, LangChain4j, JNoSQL, Introducing Lathe The Java ecosystem roundup covers framework releases, Java and Kotlin agent development kits and the introduction of a new Java language server.
🍃 Spring Updates
🧩 Spring AI Modular RAG and TypeSafe Jev: Retrieve More, Keep Only What Answers Combines query expansion with TypeSafe Jev judgments to refine retrieval results before adding relevant context to an LLM prompt.
- Two LLM calls expand and clarify the user’s question before retrieval.
- Jev evaluates retrieved chunks, helping filter context before prompt construction.
🏠 Spring AI Recipe: Running System One Models Locally Running System One models locally to make fast, focused decisions without relying on a general-purpose LLM for every judgment.
- Local models such as Nimble and Laya can handle focused decision-making tasks.
- Local execution offers an alternative to sending every decision to a remote model.
⚖️ Spring AI and TypeSafe Jev: Fast, Cheap, Structured Decisions Integrates Jev’s structured judgment API to classify, score and evaluate application state without generating conversational text.
- Typed questions allow multiple judgments against a supplied state in one call.
- Structured decisions provide an alternative to using a chat model for every evaluation.
🔍 Extra Reads
AI Tools, Models and Workflows
🧑💻 Anthropic Studied How People Actually Use Claude. The Findings Were Surprising — An analysis of Claude usage examines differences between personal and work use, requests for explanations and delegation habits.
🧰 What is Codemode — Codemode explores using executable code alongside MCP tools as an alternative way to compose and operate agent capabilities.
💸 Why Isn’t The Industry Freaking Out About DeepSeek 4.1 Flash? — A developer’s experience with DeepSeek 4.1 Flash examines its perceived capabilities, cost and competitiveness with frontier models.
Software Craft and Development
🌱 Open Source as We Know It Is Dead — Considers how AI-assisted software development may affect the sustainability and future of traditional open-source projects.
🛠️ Vibecoding isn’t as fun as writing code by hand — Why hands-on programming, learning and solving concrete problems can be more rewarding than delegating implementation.
✍️ Anti-Patterns in Software Blogging — Common software blogging mistakes offers practical guidance for writing clearer, more useful technical articles.
Infrastructure and Platform Strategy
🧾 We’re going to need default hard budget caps on pretty much everything — Hard spending limits for usage-based services and APIs can prevent costs from exceeding predefined monthly budgets.
☁️ Deno is joining Cloudflare — Deno’s move to Cloudflare aims to simplify self-hosting Workers and Durable Objects.
🔒 What good is a best-effort exclusive lock, anyway? — Database locking behavior can differ across systems.
Originally posted on marconak-matej.medium.com.