Voice agent
An AI agent that takes phone calls, holds spoken conversations in real time, and triggers actions from voice input — handling customer support, scheduling, and outbound calling.
Voice agents combine three primitives: real-time speech-to-text, an LLM for reasoning and tool use, and low-latency text-to-speech. The hard problem is latency end-to-end — every layer must respond in under 300ms or the conversation feels robotic.
In 2026 the production-grade leaders are Parloa, Sierra, and Vapi. They handle real customer-facing call flows: tier-1 support, appointment scheduling, outbound surveys, and increasingly tier-2 issues that require tool calls (look up an order, refund a charge, escalate to a human).
Buyer questions to answer before deploying: what languages do you need natively, what tools must the agent call (CRM lookup, payment refund), what escalation rules apply, and what is the SLA for the human handoff. The technology works; the deployment design is what determines whether it succeeds.
Parloa, Sierra, and Vapi are the production voice agents we track. Parloa leads on enterprise CCaaS deployments; Sierra handles end-to-end customer journeys; Vapi powers developer-built voice apps.
See voice agentsWhere this shows up
Frequently asked
How natural do voice agents sound in 2026?+
The best (Parloa, Sierra, ElevenLabs-based stacks) routinely pass "is this AI?" testing in tier-1 calls. The tells are subtle: very consistent pacing, no genuine throat-clearing, occasional over-formality.
Can voice agents handle accents and noise?+
Yes, materially better than 2024. Frontier STT models handle most accents reliably; background noise is filtered well. Heavy regional accents and code-switching remain harder than English with a clean line.
How much does a voice agent cost?+
Hosted vendor: $0.05–$0.30 per minute of conversation. Self-built on Vapi or similar: $0.02–$0.10 per minute in token + STT/TTS costs. Enterprise CCaaS deployments are per-seat or per-call with volume tiers.