“WebRTC was designed for video calls, not for AI inference. Forcing it into that role creates the latency and reliability problems OpenAI is now scrambling to fix.”
OpenAI built voice on the protocol every browser already shipped because it was the cheapest path to launch. Now they own a real-time product on top of a stack designed for two-person Zoom calls and they are paying for it in lost packets and dropped sessions. Picking the easy bus always costs you later. The bill comes due about the time your competitors notice you cannot scale.