I’ve built voice agents that hit that 80–90% human-like level in real use cases. It comes down to the right LLM, voice stack, and iterative testing. Happy to share examples—what use case are you focusing on?