Still cooking! A few more new projects in the past 2 months:
- ZIO HTTP MCP client/server library: https://t.co/QJGxjQihnQ
- ZIO Bedrock library: https://t.co/GH6ARfRUVo
- sbt revolver alternative for sbt 2.0: https://t.co/vZhsL0aLFS
- sbt webjars plugin: https://t.co/nAxHC90eAk
- sbt plugin for sass: https://t.co/ohpUVZqoNV
The past 3 months have been the most intense and fruitful of my entire career. I've been cooking, coding, vibing, architecting, designing, imagining, and delivering production systems while teaching developers around the world how to build enterprise-grade AI systems.
I'm on my way home from the amazing Spring I/O conference and reflecting on the seemingly insane number of things I've delivered / helped with over the past few months. Here are the most interesting:
- SkillsJars: Agent Skills for the JVM ecosystem. Gaining rapid ecosystem adoption for enterprise needs.
- https://t.co/UqWP34DpK0: Improvements for Java / Kotlin / Scala library Agent Experience (Valkey caching, more MCP tools, a new Scala ZIO HTTP MCP library to power it)
- https://t.co/U2LpWvO0BU: A curated AI resource list for Java & Kotlin developers (along with a generalized approach to spec driven, AI assisted websites)
- Spring AI AgentCore 1.0: The easy way to deliver enterprise-grade AI Agents & MCP servers on AWS
- acp2web: Local code assistants available anywhere via ACP
- MCP server for my Effect Oriented Programming book
- MCP server for the Spring AI book that Josh Long are working on
Along the way I presented and led hands-on Spring AI / Bedrock & MCP sessions at Jfokus, DevNexus, JavaOne, Voxxed Amsterdam, AI4J, Spring I/O, and GIDS in Bangalore next week. And I joined the Agentic AI Foundation Technical Committee, helping steer standards for the agentic world.
It has been a wild ride and I'm loving how AI has empowered me to move at a pace that a year ago was inconceivable. There is much more to come and I’m grateful for the support and collaboration with so many amazing people!
Added support to the ZIO HTTP MCP Library for the upcoming 2026-07-28 MCP release. This is a pretty big change to the transport but it didn't introduce any library consumer changes. Testament to the MCP folks to make a major change in the underlying stuff without requiring API changes!
@deusaquilus Funny thing happened... At AWS we aren't allowed to use our internal AWS accounts to expose external endpoints. So my little demo generated a SEV2 issue that was immediately escalated to my manager. Lesson learned. Going forward I'll use my personal AWS account for demos :)
What if building a Software Factory is just like building a programming language:
1. Build a compiler in another programming language
2. Rewrite the compiler in the new language
3. Use the new compiler to compile future versions of the language
This self-referential design is called a bootstrap and is an old and brilliant idea. It’s so beautifully general, it wouldn’t make sense if it wasn’t also at the core of building general AI systems.
At @alchemy_run, our goal is apply this to the problem of building a self-improving software factory. It is inherently self-referential and recursive, just like a programming language.
First, we need to design a textual representation of a software factory. This is Effect + Alchemy Resources and Bindings. This language can express composable cloud resources that we use to build things like Task Queues, Merge Queues, Review Processes, Agents, Rest APIs, Background Jobs, etc.
Then, we need a compiler that can transform this representation into a deployable service. Like building an executable. This is alchemy deploy (the iac engine). Static code in -> deployed service out.
Finally, we need a runtime. This is Effect. Type checked errors and requirements. Composable functions and layers. Production necessities like otel instrumentation.
All this comes together to form a principled, self-hosted software factory compiler toolchain. The factory represent itself and builds itself.
Prompt Engineering -> Context Engineering -> Loop Engineering -> Graph Engineering -> ??? Engineering
Y’all are so close to discovering Typed Function Programming!
@PeterB585337060@JustDeezGuy Oh! Tell me more about Lattice theory and if there is anything in that realm that an agent can write & be programmed with.
Embabel 1.0.0 is GA!!
https://t.co/UH7Uy3jSZN
Thanks in particular to @igordayen and @aheifetz for leading the charge, and to our many community contributors. Open source rocks!
Come see just how nice AI on the JVM can be.
@java@springcentral
https://t.co/14AAb8VzwJ
Skills are a great way to bring specific knowledge into your AI. But how do you manage your skills?
@JamesWard built SkillsJars to manage your skill.md files like any other dependency.
🎥 Watch the full episode: https://t.co/my1PAsuwPd
The JVM languages (Java, Kotlin, and Scala) are also great for this with fast incremental compile support. Quick iterations and strong type systems are essential for high-quality agentic development.
I agree with @ThePrimeagen here, correctness without type checking speed is no longer acceptable. That's why for most use-cases I don't consider Rust to be good (albeit loving it in some cases) and why I think TS 7 (go-based checker) is very important.
https://t.co/q6cSSFtL53