Voice agent
Also known as: AI voice agent, calling agent, outbound AI agent, phone agent
A voice agent is an AI system that uses spoken language as its primary interaction channel. The term covers two distinct directions. Inbound voice agents handle incoming calls for businesses — customer support, appointment scheduling, order status — and have been in production at scale since 2024. Outbound voice agents are the newer and more discussed side: an AI that makes calls on your behalf, navigating IVR menus (the 'press 1 for billing' phone trees), hold queues, and live conversations to complete a task and report back.
The category crossed a meaningful threshold in mid-2026 when Google and a cluster of well-funded startups began enabling outbound AI calling. For years, major assistants avoided placing calls due to safety concerns and technical difficulty. That line moved. The shift matters not just as a feature but as a signal about where agent capability boundaries are going: agents are now expected to complete tasks, not just gather information. The useful metric has moved from 'does the agent sound natural?' to 'did the appointment get booked, did the refund get issued?'
For builders, voice agents introduce infrastructure requirements that differ from text-based agents: telephony integration (SIP trunks or APIs like Twilio, Deepgram), low-latency speech-to-text and text-to-speech, and interruption handling (the model needs to stop mid-sentence when the human talks). The hardest part is not the conversation opener — it is the middle of the call, where hold music, menu trees, and unexpected questions require the agent to stay on task without losing context.