How to catch wrong answers from your AI agent that don't show up as errors in traces
A wrong answer that raises no error is a mismatch between what the agent said and something you can check: a tool result, a retrieved source, a system of record, or what the user did next.