Voice Agent Evaluation Tools Face Real-World Testing Gap

Voice agents often pass demos but fail on real calls, like transcribing confirmation codes from noisy parking garages. Evaluation must test the full pipeline, not one model. The gap between demo success and call 4,000 is the core challenge.