Agent.run_sync() is not directly patched but likely works transitively since it calls Agent.run() internally. This should be verified with a test.
pydantic_ai.direct.model_request() is a standalone LLM call API that bypasses Agent entirely. The Model.request patch in patches/model_request.py may catch the underlying call, but this also needs verification.
Both should have explicit tests confirming span creation.