Large Language Models enable powerful capabilities, but production-grade AI agents require more than prompt engineering. In enterprise environments, prompt chaining often leads to non-deterministic behavior, limited observability, and systems that are difficult to test and control.
This workshop introduces a planning-driven approach to agent design using Embabel, an open-source framework created by Rod Johnson and built on Spring AI. Instead of orchestrating behavior through prompts, Embabel applies Goal-Oriented Action Planning (GOAP), where agents are composed of strongly typed actions, goals, and domain models.
Participants will build a working agent inside a Spring Framework application, learning how to integrate LLMs as controlled components within a deterministic planning system. The resulting agents are explainable, testable, and capable of dynamic replanning when conditions change or failures occur.
By the end of the workshop, attendees will understand how to move from fragile prompt-based systems to robust, production-ready agent architectures without introducing a new tech stack or sacrificing architectural control
