Hi everyone,

One of the biggest friction points I've encountered while building Agentic AI apps in ODC is the Compilation Tax. Every time we needed to tweak a system prompt, whether to refine the tone, fix a hallucination, or adjust a rule, we had to republish the entire application.

I wanted to share an architectural pattern I’ve been using to solve this by treating prompts as Content rather than Code.

The Concept: A Headless Prompt Assembler

Instead of hardcoding prompts in logic or using the basic builder configuration, I built a standalone Prompt Assembler ODC App. It decouples the prompt lifecycle from the code lifecycle using a headless architecture:

1. Consumer Apps request prompts via Service Actions (passing only data keys, not instructions).

2. The Assembler constructs the prompt dynamically using a custom Regex-based injection engine.

3. The Benefit: We can update prompts in Production instantly without a deployment pipeline, while still maintaining strict governance and versioning.

I wrote a detailed breakdown of the architecture, including:

• How to handle JSON Output Contracts (the biggest risk with this pattern).

• The Regex logic for dynamic variable injection.

• Governance strategies for locking prompts in Production.

If you are dealing with AI deployment bottlenecks in ODC, you might find this pattern useful:

👉 Read the full architectural breakdown here: https://medium.com/@michael.de.guzman/stop-hardcoding-prompts-in-odc-19920bd6935f

I’m curious if anyone else has tackled Zero-Deployment updates for AI in ODC yet?