I’m currently building a Virtual Agent topic focused on incident creation. The flow works as follows:
- The agent collects the user’s issue and additional details.
- It performs a Knowledge Search within the topic to suggest relevant KB articles.
- If the articles aren’t helpful, the agent proceeds to create an Incident (INC).
Recently, I’ve been trying to enhance the experience by:
- Predicting the appropriate Business Offering based on the user’s inputs and displaying it for confirmation (with the option for users to change it).
- Adding a “Web Search” option in case users want to explore external resources before creating a ticket.
However, I’m starting to question whether I’m approaching this correctly. My topic feels very structured and step-driven, whereas LLM-based topics with AI Agents are meant to be more free-flowing and conversational.
Am I over-engineering this? Should I be leaning more into agentic orchestration instead of building a structured decision tree inside the topic?
I’d really appreciate any feedback, design recommendations, or best practices from those who’ve implemented similar solutions.
Thanks in advance!
