Adam Kalsey: Everyone's Building Voice AI Wrong - and You Already Know the Fix

If you've ever been trapped on the phone with an AI agent that talked over you, went dead silent for six seconds, then handed you to a human who knew nothing about the call you just sat through, then you've heard the problem. What might surprise you is that you already know the solution. You've known it for a decade. Voice AI didn't invent a new class of problem. It resurfaced an old one. Where you put the boundaries between the parts of your system matters. And a whole generation of builders who came from the web are getting it catastrophically wrong, because on the web those boundaries are free and on a phone call they cost you everything. We'll dissect a real, gloriously terrible AI call line by line. We'll do the latency arithmetic that explains why reaching for a bigger model makes things slower and worse. We'll cover the provider failover nobody builds until 3 a.m. on the night it takes them down. You'll leave this talk able to look at any voice agent -- yours or a competitor's -- and know exactly where it's going to break, and what to do about it.