Your observability tool isn't ready for agents

You're paying for two observability stacks to watch your agent. One traces the request through your handler and database, and when it reaches the model call, all it sees is a POST that took four seconds. The other has the prompt and tokens, but not the query that returned bad context.

When an agent fails quietly, nothing errors, latency looks fine, and someone on your team spends an afternoon across tabs correlating timestamps.

Pydantic Logfire takes a different approach. It's built on OpenTelemetry, so the model and the system around it belong in the same trace. We'll walk through how that works: what gets instrumented, where the data lands, and what those traces are worth once the incident is over.

A 45-min session, with time for your questions at the end.