Python: [Bug]: Duplicate LLM Telemetry Emission

[sphenry]

Description

Same gen_ai.response.id emitted across multiple spans/services, inflating token and cost accounting in Application Insights. Usage dashboards and cost assertions require manual de-dup logic; raw queries overcount by ~2x.

Workaround is to designate responsesapi as canonical source

Code Sample

Error Messages / Stack Traces

Package Versions

python-1.0.0rc4

Python Version

No response

Additional Context

No response

[eavanvalkenburg]

I have a fix up in PR #4685:

• Python: avoid duplicate agent response telemetry #4685

The change uses ContextVar-based ownership so the outer agent span stops duplicating gen_ai.response.id and token-usage telemetry that already belongs to the inner chat span, while still preserving agent-level response telemetry for agents that do not rely on a chat client.