The original a16z voice agent thesis, which now reads like a founding document for the space. The core claim: humans will only pick up the phone when the call has value to them - which means AI should handle everything else. Every inbound call to a business that doesn't require genuine human judgment is a conversation the machine should own. That's most of them.

What makes this piece edgy in retrospect is how confident it is about the quality threshold. Moore and Acharya argued that voice AI just needed to cross a "good enough" bar to collapse the economics of human call centers. That bar has largely been crossed in narrow workflows - but the piece underestimated how much trust, not just capability, would determine adoption. People hang up on AI not because it gets the answer wrong but because they didn't want to talk to a machine in the first place.

Essential context for the current moment. Read the original, then the 2025 update (Article 1 above), and you'll see exactly what the field learned between the two.