Traces contain valuable signal about an agent's capabilities and how users engage with it. Insights Agent helps you find signal into your traces by analyzing traces to discover and surface common usage patterns, agent behaviors, and failure modes.

Once you trace your data to LangSmith, you have a few options for how to categorize usage insights.

• Group by usage patterns: Cluster based on common usage patterns. This helps you understand how users are actually using your agent.

• Group by poor interactions: Cluster based on how your agent is messing up. We will look for signals in each conversation that indicate a negative interaction (user getting frustrated, etc), and then group the root causes.

• Customize configurations: Insights Agent is highly configurable. You can specify which categories it should group by, filter on existing attributes (like traces from a particular time period or keywords in a chat), define new attributes, and save configs for future use.

Insights Agent is now generally available for LangSmith Plus and Enterprise cloud customers. Check out our docs or video tutorial to get started.