In this video, I’m breaking down Google’s new Gemini 3.1 Flash Live and why I think it just changed voice agents forever. The big shift is that it’s no longer doing speech-to-text and then text-to-speech—this is straight speech-to-speech, which makes the interaction feel way more human and noticeably faster. On top of that, it can see what you’re showing it, so you can have real-time, visual conversations (like when it identified my setup and even guessed my mic correctly). I also walk through what Google is calling their biggest upgrade yet and point out the practical wins for anyone building voice agents: lower latency, better performance in noisy environments (traffic, horns, background chaos), and higher accuracy with alphanumeric strings. And because it’s interpreting audio directly instead of relying on a transcript, you get more contextual awareness—things like sarcasm, stress, or frustration actually come through. That’s a huge deal for customer support bots and sales agents where tone matters as much as the words. I also highlight a benchmark where it beats Gemini 2.5 Flash by about 19% on multi-step function calling, which is exactly the kind of capability you want for real automations.