Production agent tests should cover failure recovery
AI agent should not only check whether the agent completes the ideal task. When a run fails or stops halfway, the team should be able to see the before the failure, the that changed an outside system, and the exact data or change that was sent out.
The run should also record an or an outside receipt so repeated work can be handled safely. The should show what evidence the model used and whether the next operator can safely continue, retry, undo the effect, or stop the run.
Without that information, an agent can look good in a demo but still be weak for real . A trace should be more than a long technical log; it should become a recovery record with state, decisions, outside receipts, policy results, and ownership.
Key points
- Task success alone is not enough for agent .
- Failed runs should show the and the that affected outside systems.
- An or outside receipt helps prevent duplicate work.
- A trace should be useful for recovery, not just .
- The next operator should know whether to continue, retry, undo, or stop.