Fewer Tokens, Better Java: Spec-Driven Development at LLM Speed

LLMs were trained on Java's open ecosystem: JSRs, JEPs, normative specifications, and public code. That pretraining is an asset most developers leave unused.

In this live coding session, I will map each capability to Java packages using only public, standard notations. Then, I will let the LLM implement them. There will be no CLI, dependencies, Markdown, or sync, and there will be roughly 4,000 tokens per spec.

The result: minimal external dependencies, few turns, and a fraction of the usual inference cost.