Large Language Models enable powerful new AI capabilities, but production-grade enterprise agents require more than prompting alone. Prompt chaining often results in non-deterministic behavior, limited traceability, and systems that are difficult to control and test.

This talk introduces Embabel, an open source agent framework for the JVM created by Rod Johnson, founder of the Spring Framework. Embabel enables controllable, explainable, and production-ready agentic systems directly within existing Java and Spring applications.

Rather than orchestrating behavior through prompts, Embabel applies Goal-Oriented Action Planning (GOAP). Agents are composed of strongly typed actions, goals, and domain models, allowing execution paths to be dynamically replanned when assumptions change or steps fail. The result is agent behavior that remains explainable, deterministic, and testable.

The talk demonstrates why LLM-driven orchestration often breaks down in enterprise environments and how teams can build robust, maintainable, and controllable AI agents without introducing a new tech stack, relying on fragile prompt logic, or sacrificing architectural control.