The sixty-cent phone call.

On July 1, xAI released something called the Voice Agent Builder: a no-code tool that turns a plain-language description of how your phone calls should go into a working AI phone agent, in about two minutes.

The features are what you'd expect by now — it answers calls, books appointments on a Google or Outlook calendar, sends confirmation emails, checks order status through your systems, and hands the call to a human when it should. Every call is recorded and transcribed so you can read exactly what it said. What's actually new is the price: $0.05 per minute of audio, plus $0.01 per minute for the phone line. Each account includes a free phone number. There's no monthly platform fee. A ten-minute customer call costs roughly sixty cents.

We wrote earlier this year about why this category matters: across small service businesses, only about 38% of incoming calls get answered by a live person. The rest hit voicemail or ring out, and most of those callers simply dial the next business on the list. The case for an AI answering your phone was already solid. What just changed is who can afford one, and who can build one.

The build-or-buy line just moved

Until recently, a voice agent was something you bought as a service. Under the hood, most voice stacks stitch together three separate systems — speech recognition, a language model, and a voice synthesizer — often from three different providers, plus a platform fee on top. Every layer had its own meter. That's why "AI receptionist" services could charge real monthly money: they were doing genuinely fiddly assembly work for you.

A single-provider tool at five cents a minute compresses all of that into one bill you can do in your head. Even a busy line — say 500 minutes of calls a month — runs about $30. And because setup is a plain-language prompt plus your own uploaded documents (price sheets, service menus, policies), the assembly work largely disappears. You describe how a call should go the same way you'd brief a new hire on their first day.

How to actually try this

Start with the after-hours line. Don't route your main number to a beta product on day one. Forward calls after closing time, or on weekends — the calls you're currently losing 100% of. Anything the agent books there is pure gain, and the risk if it stumbles is a call you were missing anyway.

Write the call flow before you touch the tool. The prompt is the product. Sit down and write out: what callers usually want, what you can promise, what you can't, and when a human should take over. If you can't write that down, the agent can't do it either — and honestly, neither can a new employee.

Upload real documents, don't paste facts into the prompt. The builder retrieves from files you attach — your price list, your FAQ, your booking policy. Keep those files current and every agent that uses them stays current too.

Set the guardrails first, not after. You can put hard limits on what the agent won't do — quote custom work, discuss topics off script, read back card numbers. Decide those limits before the first call, the same way you'd tell a new hire what to never promise a customer.

Read the transcripts weekly. Every call comes back as text with a log of what the agent did. Fifteen minutes a week reading them tells you what callers actually ask — which, as a bonus, is a free market-research report you've never had.

The honest caveats

This is a beta from a company known for shipping fast and fixing in public. Phone calls are messy — bad audio, accents, people changing their minds mid-sentence — and no benchmark score guarantees your specific callers will have a good experience. The vendors' own numbers show even the best voice agents failing a meaningful share of hard calls. That's exactly why you test it on the calls you're currently losing entirely, listen to the recordings, and expand only when the transcripts earn it.

But the direction is hard to argue with. Two years ago an AI phone agent was an enterprise project. Last year it was a monthly subscription. This month it's a sixty-cent phone call and a free number. The businesses that win with this won't be the most technical — they'll be the ones who know their own operations well enough to write them down.