Evaluation set responses are diff from Team Goal Assistant responses when deployed on Goal Centre

Summary:

Inconsistent Behavior of Out-of-the-Box Teams Goals Assistant Across UI vs Agent Studio vs Evaluation Runs

We have enabled the out-of-the-box Teams Goals Assistant Agent and observed inconsistent response behavior based on where the agent is executed.

Observed Behavior

1. When the agent is accessed from the Teams Goals Center Page UI
   • The agent correctly respects the UI context, including:
     • Selected Review Period from the dropdown
     • Active tab (Performance Goals vs Development Goals)
   • Example prompt:
     “Who on my team has zero development goals?”
   • ✅ The response is accurate and aligned with the selected review period and goal type.
2. When the same agent prompt is executed from AI Agent Studio or via Evaluation Sets