On Tuesday it was the $2 AI employee. On Wednesday, xAI came for the phone lines. The company released its Voice Agent Builder in beta — a no-code platform for standing up production-grade, human-like phone agents — and priced the whole thing at $0.05 per minute of talk time, plus a cent a minute for the phone number itself. Run the math: a phone agent that never sleeps, never quits, and speaks 25-plus languages now costs about $3.60 an hour, all in.
The pitch is consolidation. What used to take three vendors and an integration project — telephony, knowledge retrieval, tool calls, guardrails, observability — ships in one stack, with more than 80 preset voices. Launch coverage says it can clone a specific voice from roughly two minutes of sample audio, and xAI claims an agent can be configured and deployed in under two minutes. The company also published benchmarks showing its Grok Voice model well ahead of Google's and OpenAI's realtime offerings (67.3% vs. 43.8% and 35.3%, by xAI's own scoring). Treat vendor benchmarks as marketing until third parties re-run them. The pricing, though, isn't a claim — it's a menu.
Zoom out and this is the second time in one week that a frontier lab repriced an entire category of work. Anthropic made autonomous digital work commodity-cheap; xAI just did the same for the most stubbornly human channel left in business: the phone call. Appointment reminders, order status, tier-one support, after-hours intake — the economics of every one of those just moved.
Our take: The story isn't "robots answer phones" — it's that voice was the last moat for millions of small businesses' most expensive workflow. At $3.60 an hour, the question flips from "can we afford a phone line that's always staffed?" to "why isn't ours?" But the same rule applies here as with every agent: garbage process in, garbage agent out. The winners will be the operators who already have their scripts, FAQs, and escalation rules written down — the same people who win every time the price of delegation collapses.
What to watch
- Voice-cloning guardrails. A tool that clones a voice from two minutes of audio is a fraud story waiting to happen. Watch how fast xAI's abuse controls get tested — and how fast regulators react.
- The incumbents' response. Call-center software is a massive market built on per-seat pricing. Per-minute agent pricing attacks the seat itself.
- Third-party benchmarks. xAI's numbers are xAI's numbers. Independent re-runs of realtime voice comparisons will tell you whether the quality matches the price.
- Small-business adoption. If your dentist's office answers on the first ring next quarter, you'll know why.
If you're building toward this, start with the stack, not the gadget: our guide to assembling a one-person AI operations team covers where voice slots in — and what to wire up first.