AI as First Class Citizens: Orchestration, Not Thinking, in ColdBox and Lucee
AI is increasingly embedded directly into handlers, services, and core workflows, often being used to make decisions that should remain deterministic. While this approach may work for demos, it frequently leads to fragile production systems that break when AI is slow, unavailable, or incorrect.
This session focuses on a safer and more sustainable approach: treating AI as a first class citizen in ColdBox and Lucee applications without allowing it to “think” or control business logic. Instead of embedding AI directly into request paths, we will explore how AI can be orchestrated intentionally using ColdBox Custom Interception Points and the asyncManager to run AI workflows asynchronously and event-driven.
Using two real-world examples, the talk demonstrates how AI can be applied as a service layer for validation, scoring, and content assistance. One example shows how AI can analyze support tickets in the background to detect spam, prompt injection, or low-trust content, producing trust and confidence scores that drive deterministic routing such as quarantine queues, review queues, or trusted workflows. Another example illustrates how AI can enhance listing quality and SEO asynchronously without ever blocking core save operations.
Throughout the session, we emphasize clear architectural boundaries and repeatable patterns that keep applications stable and testable. AI is used to generate signals, not decisions. Domain services remain authoritative, handlers orchestrate workflows, and AI services assist without becoming points of failure.
Attendees will leave with a clear mental model and practical ColdBox patterns they can apply immediately, guided by four core principles: AI is a service, not a shortcut; signals over decisions; orchestration over intelligence; and stability beats cleverness.
Srikanth Madishetti
