Pinned Tweet
Project Babylon/HAT brings GPU acceleration to Java: LLM inference in Java, kernels approaching native performance. A major step toward a JVM that dynamically targets CPUs and GPUs efficiently without writing kernels in CUDA/OpenCL. youtu.be/0-1IGpVrFLQ
Building Project Valhalla is like building a cathedral: craftsmanship, ingenuity, and collective dedication. A single “value” keyword carrying a decade of engineering. The first preview in JDK 28 as a milestone, with more work ahead. Kudos to the team!
youtu.be/D_AmK8-UnZg
Virtual threads, structured concurrency, and scoped values simplify Enterprise AI agent orchestration—enabling scalable concurrent calls to models, tools, and services with clearer task lifetimes, cancellation, context propagation, and observability.
youtu.be/oS01pSRWKmg
#AI
Java—The Year of the Runtime! As AI accelerates coding, the JVM matters more than ever: compatibility, integrity, JIT compilation, performance, memory management, and JFR telemetry. AI Enterprise software will remain secure, robust, efficient on the JVM.
youtu.be/fXBGBovpH88?si=3ZoG…
Bernard Traversat retweeted
#LangChain4j 1.22.0 is out!
Highlights:
- OpenAI Decisions API support
- AWS Bedrock batch inference
- Automatic prompt caching for Anthropic
Thanks to all contributors!
github.com/langchain4j/langc…
Leyden AOT is taking center stage at Devoxx BE! No code changes—just blazing-fast startup, fast warmup to peak performance, and the C2 JIT continuing to push performance to the ultimate limit :-)
youtu.be/cbSl8UVdMVg?si=a7VN…
"Java for the AI world"
Oracle’s Java extension for VS Code 27.0.0 is here! 7.2M downloads and counting. New NetBeans 31, full JDK 27 support, lambda breakpoints, Java notebook fixes, and dependency updates.
marketplace.visualstudio.com…
#java #VSCode
JDK 28 FFM allocations will be much faster: Arena.ofConfined() can reuse native-memory pools without source changes while preserving lifetime, confinement, and zero-initialization guarantees. Early microbenchmarks show up to 18x improvements!
inside.java/2026/10/05/confi…
#java #FFM
Helidon combines Java virtual threads for a simple request-per-thread model and compile-time generated APIs with LangChain4j integration for models, tools, workflows. A fast, robust, secure and cost-efficient foundation for Enterprise agentic AI at scale.
youtube.com/watch?v=ewSwIZvN…
JDK 27 delivers performance gains from 2,300+ commits: Compact Object Headers on by default, faster HashMap copies, text and URL improvements, crypto intrinsics, vectorization, GC fixes and more :-) inside.java/2026/09/28/perfo…
"Java for the AI world"
A Deep Dive into SPECjbb2015 MultiJVM Mode and Tail Latency!
norlinder.nu/posts/The-Limit…
"Java for the AI world"
#java #JVM #GC #SPECJBB
JEP 546, Adaptive Heap Sizing for ZGC, is now a Candidate! ZGC will adapt heap size to application demand and available host CPU/memory—reducing manual tuning while balancing throughput, latency, GC CPU, and neighboring workloads. openjdk.org/jeps/546
#Java #ZGC
Helidon 27 is the first release under JDK "Tip-and-Tail" model—bringing the latest Helidon capabilities on the latest JDK 27. It adds full Declarative APIs, MCP, LangChain4j, JDBC, and Scoped Values support for virtual threads.
medium.com/helidon/helidon-2…
JEP 545, Faster Startup and Warmup with ZGC, is now a Candidate! Complementing Leyden AOT, ZGC reduces the trade-off between fast startup with a small initial heap and smoother performance with a large, pre-touched heap😀
openjdk.org/jeps/545
#java #ZGC
JDK 27 strengthens Java security with TLS 1.3 post-quantum hybrid key exchange, a refined PEM API, faster cryptography, updated trust controls, TLS certificate compression, and runtime security-property visibility via jcmd. seanjmullan.org/blog/2026/09…
#java #PQC
JavaFX 27 is now available! Metal is the default rendering pipeline on macOS, with rich-text and accessibility improvements. Build enterprise desktop experiences for the AI world, where people can review AI insights, resolve exceptions, and stay in control oracle.com/java/technologies…
Java 27 available now! 9 JEPs improve productivity, performance, memory footprint, serviceability, and post-quantum TLS. With ~2,300 enhancements and fixes, it strengthens Java for the Enterprise AI agentic world :-) oracle.com/java/technologies…
JEP 544 is now Candidate! A major milestone for Project Leyden and the C2 JIT compiler, it speeds startup and warmup by moving compilation into a training run while preserving C2’s dynamic optimization and peak performance 🙂
openjdk.org/jeps/544
I would add “Integrity by Default” to that wonderful list as well. It reflects a deliberate move toward a safer, more reliable Java platform—one in which strong encapsulation and secure defaults are the norm, especially for the age of GenAI :-)
adambien.blog/roller/java_wa…
#java #ai



