Coding agents, runtime improvements and better engineering systems are changing software delivery.
🎯 Featured Article
📈 Spotify More Than Doubled Monthly Code Changes as AI Accelerated Development The shift is prompting changes to testing, monitoring, rollout safeguards and capacity planning across its production systems.
- Higher change volume exposed risks in content processing, fleet changes, compute capacity and mobile quality.
- Spotify is strengthening safeguards and quality measurements to manage changes across nearly 3,000 services.
⏩ TL;DR (Quick Recap)
- AI-assisted development is increasing code throughput, making testing, rollout safety and operational capacity more important.
- Coding agents can benefit from IDE-grade code intelligence and behavioral harnesses, especially with smaller models.
- JVM work continues across runtime performance, observability, language projects, and Java’s role in AI applications.
- Spring AI examples show ways to add fact-checking, browser-based interfaces and rich MCP tool experiences.
☕ JVM Corner
🦀 Kotlin Keeps Up With Rust Porting 37signals’ Campfire to the JVM in a day and what DHH’s benchmark table actually measures.
- The porting exercise compares a byte-for-byte implementation on the JVM.
- Benchmark interpretation matters when comparing Kotlin and Rust performance.
🧠 IntelliJ IDEA’s Java and Kotlin for Cursor and VS Code Hits 1.0 The stable 1.0 release brings IntelliJ IDEA’s Java and Kotlin intelligence to VS Code, Cursor and compatible editor workflows.
- The LSP-powered extension moves project-aware JVM intelligence beyond IntelliJ IDEA.
- The release follows a ten-week preview and supports developers using different editors and agents.
🛠️ We Didn’t Want to Build Another Java Server New Java server approach focuses on small deployments, performance and sharing application models across client and server.
- The design targets microservices and serverless workloads where startup and resource budgets matter.
- Codename One can share entity classes and validation rules between the app and server.
🤖 How Does Java Fit Into the World of AI — and How Can It Become an Even Better Fit? Java’s current AI suitability and platform improvements spanning language features, libraries and runtime capabilities.
⚙️ Java — The Year of the Runtime! Why JVM performance, observability, garbage collection and usability matter as AI agents make runtime behavior more central.
🍃 Spring Updates
🧾 Spring AI Recipe: Fact-Checking with a System One Model System One model to fact-check AI responses with focused, typed judgments instead of relying on another LLM as judge.
- Typed judgments provide a focused way to evaluate responses.
- The approach targets lower latency and cost than using another LLM for every check.
🖥️ Spring AI Recipe: Adding a Web UI with AG-UI and CopilotKit Spring AI application to a browser-based conversational interface using AG-UI and CopilotKit.
- The web UI moves interaction beyond a terminal-based chat.
- AG-UI and CopilotKit provide the interface layer for a more polished chat experience.
🧩 Spring AI Recipe: Rendering Rich UIs with MCP Apps Spring AI MCP server can return interactive interfaces alongside tool results instead of limiting responses to plain text.
- MCP Apps can present rich interfaces alongside agent tool output.
- The pattern extends an MCP server’s response beyond text-only results.
📊 Spring Boot JPA vs JDBC: How Much Can a Small App Save? A small Spring Boot benchmark found that JDBC used about 90 MiB less and reached its dashboard in roughly half the time of JPA and Hibernate.
🔍 Extra Reads
AI Agents and Engineering Systems
🎬 How Netflix Made AI Coding Agents More Efficient with JetBrains IDE Intelligence Netflix integrated JetBrains’ headless code-intelligence engine with Claude Code and Codex.
🧪 Are Harnesses Worth the Cost? Behavioral harness improved a cost-optimized model’s success rate from 3/10 runs to 10/10.
🔀 Redesigning source control at Anthropic An in-house version control system in five months after merge volume made source control a development bottleneck.
Models, Security and Web
⚡ Introducing Claude Haiku 5.5 Anthropic presents a small model for fast, high-volume tasks such as summaries, classification, database queries and coding subagent work.
🔐 Passkeys are a trap. Do not use them. This opinion piece questions passkey adoption and focuses on concerns for non-technical web users.
🕸️ Scraping websites and bot-protection Website bot-protection levels and countermeasures encountered while researching family history.
Code, Data, and Production Performance
📖 How to read code How developers approach code comprehension.
🏗️ The system around the code Deployment, operations and organizational systems often determine whether an AI-built prototype can reach production.
🗄️ Postgres Isn’t Slow. Your Storage Is A benchmark examines how storage affects PostgreSQL performance.
Originally posted on marconak-matej.medium.com.