Writing

What we have worked out about building voice agents that hold up outside a demo — cost models, latency budgets, escalation design, and the constraints nobody mentions until you hit them.