ClueCon Weekly with Adam Kalsey [Sn.15 Ep.17]: Voice AI Decisions: Observability, Compliance, & Cost

ClueCon Weekly with Adam Kalsey [Sn.15 Ep.17]: Voice AI Decisions: Observability, Compliance, & Cost

In this episode of ClueCon Weekly, host Jon Gray is joined by Adam Kalsey (VP of Product at SignalWire) to break down a question teams keep running into when building voice agents: do you go with a traditional voice AI pipeline (STT → LLM → TTS), or a newer speech-to-speech / audio language model approach?
​
They dig into the real-world trade-offs—latency vs. control, simplicity vs. observability, and why “one API call” can be great for prototyping, but risky if you need things like audit trails, redaction, deterministic business logic, or regulated-industry compliance.
​
You’ll also hear where speech-to-speech shines (like capturing tone and emotional nuance) and why many production systems land on a hybrid approach—using models where they add value, and code where you need determinism. Plus: a practical warning about over-optimizing latency so much that the agent feels unnatural on a call.