Builders of vertical AI phone agents need a way to test calls at scale because each use case takes months of manual prompt tuning

Teams deploying AI voice agents for a specific business (drive-through ordering, clinic booking, claims intake) hit the same wall: the agent only works after months of manual prompt tuning per use case, and there is no good way to simulate hundreds of difficult callers and score reliability before going live. The job an AI agent must do is auto-generate realistic adversarial test calls, run them against the live agent, and score where it breaks. Demand is verified by Hamming's Launch HN at 129 points where commenters immediately wanted to point it at their Retell agents, plus the recurring Leaping thread noting big companies spend months tuning one use case. This sits at the use-case boundary of testing a deployed vertical agent, and the gap is domain-specific scenario libraries per vertical rather than generic call recording.