How can a pilot reveal that the agent needs a different action boundary?

An insurance operations reviewer assigns case CLM-2187 to an agent in a read-only pilot. The agent finds the settlement assessment and approval but cannot find the required specialist response. It records the gap and prepares the available evidence for review.

The reviewer then asks for the missing response to be obtained. In the current environment, the agent has no action for creating a follow-up request. The case changes from ready for assembly to awaiting additional support, but no work item reaches the specialist queue. The consequence is a valid gap report with no mechanism to progress it.

This pilot result identifies a scope boundary rather than proving that the agent failed. A subsequent scope could introduce a controlled `create_evidence_request` action, a designated recipient role and a review step before submission.

The lesson is to distinguish evidence diagnosis from workflow execution. Buyers can use the first run to decide whether they need retrieval tasks, request-management tasks or both, and define acceptance criteria for each responsibility separately.