OpenTelemetry has become the standard for observing modern systems.
But when you start building AI applications, traditional traces aren't enough.
You don't just want to know that a request happened.
You want to know:
Which model generated the output?








