Spring Boot 4.2.0-RC1 is out: faster startup in executable jars, a properties migrator that can fail startup on warnings, and a new Ubuntu 26.04 buildpacks builder. Test it before GA. https://spring.io/blog/2026/10/23/spring-boot-4-2-0-RC1-available-now/
Dan Vega
Dan Vega
@danvega.dev
π Spring Developer Advocate @Broadcom βοΈ Java Champion πΉ YouTuber http://bit.ly/3zJb93b ποΈ Podcast http://bit.ly/3Tzi0CO π€ ByteSized AI http://bit.ly/3XYv87j
1.9K followers512 following262 posts
Spring AI 2.1.0-RC1: mark a tool with requiresConfirmation = true and the tool loop pauses until your app approves or denies the call. That's the guardrail you want before an agent touches real data. https://spring.io/blog/2026/10/23/spring-ai-2-1-0-RC1-available-now/
New Claude API version: 2026-10-21. temperature, top_p and top_k are gone. Send them and you get a 400, even at default values. If you set temperature in Spring AI options, check this before you upgrade. https://platform.claude.com/docs/en/api/migration-2026-10-21
Agent Skills now lives at the Agentic AI Foundation, next to MCP. 70+ agent platforms support it. One skill for your Spring conventions can work in Claude Code, Copilot and Cursor. https://www.anthropic.com/news/donating-agent-skills-to-the-agentic-ai-foundation
Oracle shipped an emergency Java fix for CVE-2026-91403 (CVSS 9.1). The JSSE flaw lets a network attacker pass forged TLS certs as trusted. Java 8 through 27 are listed. Patch your JDKs today. https://blogs.oracle.com/security/security-alert-cve-2026-91403-released
Jackson 3 rejected a null that my Boot 3 app had quietly turned into zero. I can see the case for the stricter default. But existing clients may rely on the old behavior. Would you preserve that contract or ask clients to change? https://youtu.be/HiPkoGTeNJc
What actually breaks when you move to Spring Boot 4? I upgrade a real app from 3 to 4 on camera. No slides. Just fixing compiler errors, broken tests, and changes that don't trigger a warning. https://youtu.be/HiPkoGTeNJc
Tim Ferriss won't let AI write his next book. His reason: βIf I do that, I'm done.β He still uses it for everything around the writing. Developers need the same line. Pick the skills you want to keep in your head. Do those by hand. Hand off the rest. https://youtu.be/X9nfEE8DJeI
I made a theme for IntelliJ. Kiln is warm charcoal and cream with a clay accent and calm syntax colors. Dark and light, built on the new Islands look. It's an early release, and I'd love feedback before it goes on the Marketplace. https://github.com/danvega/kiln-theme
Fetching JSON from an API and mapping it to records in plain Java. No Jackson. No Gson. No dependencies. Just HttpClient and the new JSON API coming in JDK 28 (JEP 540). π https://github.com/danvega/native-json-demo
Spring AI 2.1.0 M1 is out. What stands out to me is support for OpenAI's Responses API. https://spring.io/blog/2026/09/25/spring-ai-2-1-0-M1-available-now
Huge Thank you to @dthompsondev.bsky.social for joining us on Spring Office Hours this week. It was a fun conversation and Danny is what we should aspire to be. Great person and can't wait to catch up with him at Commit Your Code in a couple of weeks! www.youtube.com/watch?v=Jzbx...
This week's newsletter is out. New videos on Docker Compose and Spring AI guardrails, plus a first look at the Spring Boot 4 crash course I'm working on. Be the first to know when the course is released. Subscribe: www.danvega.dev/newsletter/
Embabel 1.0 is here. Rod Johnson's AI agent framework for the JVM hit GA: stable APIs, everything on Maven Central, RAG promoted to production, new model integrations, and better MCP citizenship. Builds on Spring AI, Spring Boot 3.5, Java 21. www.danvega.dev/blog/embabel...
Spring AI 2.0 moved the tool calling loop into the advisor chain. That means you can hook into it and watch it run. I built a streaming chat where every tool call shows up live as a pill, "Calling getWeatherForecast...", using one small custom advisor. github.com/danvega/tool...
Can you use Java for AI? Yes. And if you're building AI into real applications instead of training models, it might be the better choice. The training era belonged to Python. The integration era looks a lot like the JVM. www.danvega.dev/blog/can-you...
Spring AI advisors are AOP for your LLM calls. Do something before the request, something after. I built two custom advisors: one that logs which tools are visible to the model, and one that tracks token usage per call with a running total. Full walkthrough here: youtu.be/WNJFR-_N9bc
Register 47 tools in Spring AI and every one ships its JSON schema on every request. That's ~5,000 prompt tokens before the user asks anything. The Tool Search Advisor fixes it: the model sees ONE search tool. Demo + walkthrough π (Video Coming Soon) github.com/danvega/spring-ai-tool-search
Spring Boot 4.1 added a property: spring.datasource.connection-fetch=lazy Transactional methods grab a connection the moment they open, even when they never query. Cache hit, early return, slow API call. That connection sits idle. Lazy waits for a real statement. github.com/danvega/jdbc...