What is an AI voice agent?

An AI voice agent is software that conducts a phone conversation autonomously: it speaks with a synthesized voice, transcribes the other party in real time, decides what to say next using a large language model, and takes actions like booking a meeting or logging a disposition. For sales, it places outbound calls (cold or follow-up) and answers inbound calls without a human on the line.

How does an AI voice agent place a call?

The platform dials through a telephony provider (SIP/PSTN), streams the live audio into a speech-to-text engine, passes the transcript to an LLM that generates the next response, converts that text back to speech with a text-to-speech voice, and plays it on the call — looping every turn with sub-second latency. Calls are typically recorded and transcribed for review.

What does it cost to run an AI voice agent?

Costs stack across telephony minutes, speech-to-text, text-to-speech, and LLM tokens — typically a few cents to ~$0.15 per minute when assembled from raw APIs, before you add a CRM and orchestration. Bundled platforms price per included minute so the per-call cost is predictable.

Are AI voice agents legal for cold calling?

Yes, with conditions. The FCC treats AI voices as artificial under the TCPA, so marketing cold calls generally need prior express written consent, an upfront AI disclosure, DNC scrubbing, and calling-time limits. See our guide "Is AI cold calling legal?" for the full breakdown.

Guides · Product

AI voice agents for sales

An AI voice agent holds a real phone conversation on its own — it speaks, listens, decides, and acts. For sales teams that means software that can place outbound calls and answer inbound ones, qualify a lead, book a meeting, and write the result into your CRM, without tying up a rep. Here’s how it actually works and what it takes to run one well.

How a call actually flows

Every turn of the conversation runs the same loop, fast enough to feel natural:

1. Dial — the platform places the call over a telephony provider (SIP/PSTN).
2. Listen — live audio streams into a speech-to-text engine that transcribes the prospect in real time.
3. Think — the transcript goes to a large language model that decides the next line, guided by your script and guardrails.
4. Speak — a text-to-speech voice renders the reply and plays it on the call.
5. Act — at the end, the agent logs a disposition, books a calendar slot, or hands off — and the recording and transcript are saved.

The stack you need

Running an agent from scratch means assembling — and paying for — five things: telephony (numbers + SIP), speech-to-text, text-to-speech, an LLM, and a CRM to hold the contacts and results. Plus the orchestration glue and the compliance layer that keeps you on the right side of the TCPA. That’s the build-it-yourself path (think Twilio + Deepgram + a voice API + a CRM + your own code). Bundled platforms ship all of it as one product. See Lead Friendly vs Twilio and vs Retell for the trade-offs.

Where AI agents fit in an outbound motion

• Speed-to-lead — call inbound web leads in seconds, any hour, before they go cold.
• List qualification — work a large list to separate interested prospects for human reps.
• Follow-ups & reminders — confirm appointments, re-engage no-shows.
• Inbound answering — never miss a call; route or book automatically.

Deploy it compliantly

Because the FCC classifies AI voices as artificial under the TCPA, a sales agent isn’t a pure engineering project — it’s a regulated one. Before you scale, lock down consent, the artificial-voice disclosure, DNC scrubbing, calling hours, and recording-consent rules. Our TCPA compliance checklist walks through each one.

An AI agent + CRM + telephony, in one

Skip the five-vendor assembly. Configure an agent, buy a number, and start dialing — compliance built in.

See pricing

General information, not legal advice.