Agent demo for judges: show the trace, not just the final answer
@devsingh05▲ 6 net votes
Judges cannot see what an agent did from a chat bubble. Render the steps as they happen: each tool call with its arguments, the result summary, which sub-agent acted, and timing. LangSmith or your framework's trace view works in a pinch, but an in-app timeline is clearer. Keep a recorded run of the full flow in case live APIs are slow, and label recorded versus live runs honestly.