diff --git a/py/src/braintrust/integrations/claude_agent_sdk/test_claude_agent_sdk.py b/py/src/braintrust/integrations/claude_agent_sdk/test_claude_agent_sdk.py
index 6fe29edd..8e1d53b8 100644
--- a/py/src/braintrust/integrations/claude_agent_sdk/test_claude_agent_sdk.py
+++ b/py/src/braintrust/integrations/claude_agent_sdk/test_claude_agent_sdk.py
@@ -166,6 +166,8 @@ async def calculator_handler(args):
     assert task_span["output"] is not None
 
     assert result_message is not None, "Should have received result message"
+    assert getattr(result_message, "result", None) is not None
+    assert task_span["output"] == result_message.result
     if hasattr(result_message, "num_turns"):
         assert task_span.get("metadata", {}).get("num_turns") is not None
     if hasattr(result_message, "session_id"):
diff --git a/py/src/braintrust/integrations/claude_agent_sdk/tracing.py b/py/src/braintrust/integrations/claude_agent_sdk/tracing.py
index 4dea63d6..398d66f8 100644
--- a/py/src/braintrust/integrations/claude_agent_sdk/tracing.py
+++ b/py/src/braintrust/integrations/claude_agent_sdk/tracing.py
@@ -585,6 +585,7 @@ def __init__(
         self._task_order: list[str | None] = []
 
         self._final_results: list[dict[str, Any]] = []
+        self._result_output: Any | None = None
         self._task_events: list[dict[str, Any]] = []
 
         _thread_local.tool_span_tracker = self._tool_tracker
@@ -609,7 +610,11 @@ def add(self, message: Any) -> None:
             self._handle_system(message)
 
     def log_output(self) -> None:
-        """Log the last accumulated assistant message as the root span output."""
+        """Log the canonical root span output for the request."""
+        if self._result_output is not None:
+            self._root_span.log(output=self._result_output)
+            return
+
         if self._final_results:
             self._root_span.log(output=self._final_results[-1])
 
@@ -715,6 +720,11 @@ def _handle_result(self, message: Any) -> None:
             ctx = self._get_context(None)
             if ctx.llm_span and (usage_metrics or usage_metadata):
                 ctx.llm_span.log(metrics=usage_metrics or None, metadata=usage_metadata or None)
+
+        result_value = getattr(message, "result", None)
+        if result_value is not None:
+            self._result_output = result_value
+
         result_metadata = {
             k: v
             for k, v in {