Intents in Agent Optimization
Key Takeaway
Intents are not real time: Agent Optimization generates them only after a session closes. A cluster tag forms only when at least 10 intents are similar enough in meaning, and the trace lists only tagged intents.
An intent is a set of interactions within an agent session that address one specific user request, such as a cancellation request. Agent Optimization creates intents from closed sessions and groups similar intents under shared cluster tags. Intents help you understand what users are asking and how your agent responds. Reviewing them shows you what users ask most and how well the agent answers, so you can analyze interaction patterns and improve agent performance.
How Agent Optimization creates intents and cluster tags
Agent Optimization runs two pipelines in order. First, the intent pipeline analyzes closed sessions and uses an LLM to generate intents. Then the clustering pipeline groups intents from many sessions by shared user goal and assigns cluster tags, such as Cancellation Request or Order Status.
With cluster tags, you can study patterns at the tag level instead of looking at one raw intent at a time.