In the rush to integrate Large Language Models (LLMs) into technical writing, many organizations are making a fundamental architectural mistake: they are letting their CMS vendor decide how the AI interacts with their content. When automation is a “black box” feature, you lose the ability to audit, refine, or pivot your strategy as AI models evolve.

At SiteFusion ProConsult, our technical architecture is built on the belief that workflow logic must be separated from content logic. We use Camunda as a native BPMN workflow engine because it provides a visual, standardized map of every process—from initial authoring in Fonto to final delivery via MarkLogic.

When you move toward AI-driven pipelines, this orchestration layer becomes your primary control mechanism. If you want to insert an AI step to check for JATS XML compliance or generate metadata, that shouldn’t require a custom code deployment from your vendor. It should be a task you can model and modify within your BPMN diagram.

As we prepare for ConVEx Pittsburgh next week, I’m looking forward to discussing why “process orchestration” is the actual differentiator in an enterprise CMS. If your workflow is hard-coded, you aren’t agile; you’re just automated.