Modern Java with AI
Summaries, reflections, and practical takeaways from my live coding sessions. I share useful patterns, failed attempts, and what changed my mind while building with Java and AI.
10 posts published
A live session with Adam Dymitruk where we separated Event Sourcing from Event Modeling, sketched a first model in Miro, and generated a running Java demo with Copilot in Docker sbx, even though we had limited time for a deep code walkthrough.
Simon Martinelli introduced Spec-Driven Development and the AI Unified Process, then walked through all four phases live — from Vision document to a running Vaadin app scaffolded from a Maven archetype, with Claude Code handling most of the implementation.
Ralf D. Müller introduced Semantic Anchors, Semantic Contracts, and Spec-Driven Development — a structured path from a rough idea to a full architecture document, specification, and GitHub Issues, without writing a single line of code.
Kevin Wittek and I ran a deliberately vulnerable Java project through Docker Sandbox to find out whether sbx can actually contain an AI agent in YOLO mode. The isolation held, and it changed how I think about running agents safely.
We used IBM Bob's 40-coin trial budget to build a JavaFX chess game from scratch. The prompt enhancement and custom modes impressed, but we ran out of tokens before the tests could be reactivated.
This live session walked through five levels of AI coding autonomy in Java, but the real takeaway was simpler: sustainable AI development depends on small reviewable steps, explicit specs, and deterministic quality gates.
A two-hour live refactor that moved a file search app from PostgreSQL to Neo4j using GraphRAG and LangChain4j. The session didn't land, but four lessons made it worth the attempt.
What I took away from this session with Tim te Beek: OpenRewrite becomes especially powerful when AI is used to help build deterministic, testable recipes instead of making uncontrolled code changes.
Two predefined prompts, one shared Java app, and strict CI gates: the one-shot version started but failed when retrieving results, while the split-prompt workflow produced the more stable app and better quality metrics.
This live session made a strong case for using AI as a disciplined implementation engine: define constraints first, separate planning from coding, and review output against architecture instead of accepting it at face value.
