Vercel CLI lets AI agents find, price and buy domains with human sign-off

Vercel has extended its command-line tool so AI agents can carry out the full domain purchasing workflow, from discovering candidates to initiating the transaction. According to a post on the company's changelog, the agent handles everything except the step that matters most: approving the payment.

How the agent workflow runs

The capability is built around a short chain of CLI commands. vercel domains search lets the agent discover candidate names, vercel domains check reports whether a name is available, and vercel domains price returns registrar pricing. Once the agent settles on a candidate, it initiates the purchase with vercel domains buy.

Vercel suggests a prompt as simple as asking an agent to buy a domain for a site, optionally seeding it with a name fragment such as "shipsf26". From there, the agent works through the sequence on its own.

Approval is a hard boundary

The spend decision deliberately stays with the user. When vercel domains buy runs non-interactively, it does not complete the purchase. Instead, it returns a structured, machine-readable error along with suggested follow-up commands, so the agent hands the confirmation back to the human. That rules out unauthorized spending by construction rather than by prompt instruction.

Whoever approves the purchase can review the price, the auto-renew setting and the registrant contact details, then confirm them either in an interactive terminal session or through the Vercel dashboard.

A bundled skill teaches agents the sequence

The CLI also ships with an agent-facing skill covering the discovery, pricing and purchase workflow. Vercel says this means agents follow the full sequence without being told each individual step, and points to its documentation for getting started.

Why it matters

This is a small feature with larger significance: agentic commerce in production, with an AI agent executing a real paid transaction end to end rather than only drafting content or writing code. Domain registration is a sensible first product for the pattern — purchases are standardized, prices are visible before checkout, and the damage from a bad buy is bounded.

The design also offers a template for how tools should expose money-moving operations to agents. Rather than trusting a model to obey an instruction not to spend, the tool makes approval a structural boundary: the non-interactive command simply cannot complete the payment. As agents take on more operational work — provisioning infrastructure, renewing subscriptions, managing third-party services — that split between agent-executable steps and human-gated commitments looks likely to become a standard interface convention.