AI phone agents now cost pennies a minute. Is it time to answer every call?

Every missed call is a customer who might dial your competitor next. For most small businesses, answering every call has meant hiring another person, paying an answering service, or living with voicemail.

Two launches in the past five weeks make a fourth option far more practical: an AI agent that answers the phone, handles the routine calls, and passes the rest to your team.

What happened

On September 10, OpenAI made GPT-Live-1 available to developers. It’s a “full-duplex” voice model, which means it can listen and speak at the same time, so callers can interrupt it the way they’d interrupt a person. It can answer a phone, and voice sessions cost $0.05 per minute, billed by the second. The model that does the thinking behind the conversation, and any tools it uses, are billed separately.

On October 14, xAI released Grok Voice 3.0 at $0.08 per minute ($4.80 an hour). xAI says Starlink’s call centers use it to answer 20,000 customer calls a day in multiple languages, and that “roughly 70%” are resolved with no human involved. That’s the company’s own figure about its own product, so treat it as a best case. Still, it shows how far voice agents have come.

What a phone agent actually costs

The per-minute price is the easy part. At those rates, 1,000 minutes of conversation comes to $50 to $80 in voice fees. A real phone agent also needs:

• the model behind the voice, if it’s billed separately, to look things up and decide what to do;
• a phone number and calling service to connect the agent to the phone network;
• connections to your systems, like your calendar, scheduling software, or CRM, so the agent can book an appointment instead of just taking a message;
• setup and testing, which is where most of the work is.

Even with all of that, the running cost of a well-scoped agent is modest next to a lost customer. The bigger investment is designing it well.

Where phone agents fit

Good first jobs for a phone agent:

• After-hours and overflow calls. Answer, find out what the caller needs, take their details, and text or email your team right away.
• Routine questions. Hours, directions, services, what to bring, and how to pay.
• Booking and rescheduling. Put appointments straight into your scheduling system, and confirm them.
• Intake. Collect the details your staff will need before anyone calls back.

Poor fits, at least to start: upset customers, emergencies, sensitive conversations, and complicated sales. Those should reach a person quickly, and a good agent knows to send them there.

How to set one up well

1. Start with one job. After-hours calls are a common first step, because almost anything beats voicemail.
2. Write down the rules. What the agent may answer, what it must hand off, and how fast a person follows up.
3. Be upfront. Let callers know they’re talking to an AI assistant. It builds trust.
4. Give it your real information. Hours, services, prices, and policies, kept up to date in one place.
5. Keep a person in the loop. Every call should leave a transcript or summary your team can see.
6. Measure it. Track missed calls, bookings, and handoffs before and after, so you know what it’s worth.

The bottom line

Per-minute pricing makes it cheap to try a phone agent on one slice of your calls and see what happens. The technology is ready for routine calls. What separates a helpful agent from a frustrating one is the setup: clear rules, good information, and a fast handoff to a person.

That’s the kind of work our AI agents service covers, for phone, website, and back-office agents.

Sources

• OpenAI: GPT-Live-1 model documentation
• DataNorth: OpenAI launches GPT-Live-1 in the API, September 11, 2026
• xAI: Introducing Grok Voice 3.0, October 14, 2026