A practical voice AI system is a loop, not a single agent.
Audio comes in, partial understanding forms, tools run, state changes, and the UI needs to keep the person confident that something coherent is happening.
Talk outline
- What changes when the interface is real-time.
- Where agents help and where they add unnecessary motion.
- Why evaluation has to include latency, correction, and recovery.
- How product feedback loops should shape the agent architecture.
Takeaway
Voice AI should be designed from the product loop outward. The model is one part of the loop, not the whole product.