Everyone is jumping into AI voice right now. But most systems still struggle in real conversations. The biggest issue I keep seeing is latency. A lot of tools rely on speech to text, then text to speech. That delay adds up and makes conversations feel unnatural, especially in sales or support use cases. Lately I have been testing a different approach using voice to voice architecture instead. The difference is noticeable: - Responses feel almost instant - Conversations flow more naturally with interruptions - Lower cost compared to multi step pipelines - Handles higher call volume more reliably It actually starts to feel less like a bot and more like a real interaction. Curious to hear from others here: Have you tested AI voice agents in real use cases?What has been the biggest bottleneck for you so far, latency, cost, or reliability? Would be interesting to compare notes.