What it does

Topics are the scripted conversation paths in an agent. Each has trigger phrases that fire it, and a graph of nodes — messages, questions, conditions, variables and actions — that control the exact flow.

Key facts

• Trigger phrases train the intent model; give a spread of real phrasings, not one canonical sentence.
• Nodes cover messages, questions, conditions/branching, variables, and calls to actions or other topics.
• System topics (greeting, fallback, escalate, end of conversation) ship by default and are editable.
• Variables can be topic-scoped or global, and global variables carry state across the whole conversation.

When to use / skip

Use topics when you need deterministic, auditable flows — compliance scripts, structured intake, anything that must happen the same way every time. Skip heavy topic-building for open Q&A; generative answers and orchestration handle that with far less maintenance.

Configuration decisions

• Which journeys must be deterministic (topics) vs handled by orchestration.
• Global vs topic variables, and what state needs to persist.
• How trigger phrases are curated and reviewed over time.

Gotchas

• Overlapping trigger phrases across topics cause the wrong topic to fire. Keep intents distinct.
• Big branching topics get unmaintainable quickly — break them into callable sub-topics.

Consultant notes

• In generative orchestration, topics become tools the model can call; write descriptions accordingly.
• Deterministic flows are still the right call for regulated steps — don't let the model improvise a compliance script.
• Version topics through solutions; ad-hoc edits in production bite you.

---

Revisit if orchestration changes how topics are invoked.