Agents Got Cards

Mercury announced agent-issued corporate cards this week. Real virtual cards. Spending limits, merchant controls, automatic receipt capture. An AI agent books a flight or provisions a cloud server and the transaction shows up in the same ledger as every other company expense.

This is not a fintech gimmick. It solves an actual operational bottleneck. Right now, agents that need to spend money either share a human’s card — which is an audit nightmare — or get blocked entirely. Most companies default to blocking. The agent can research vendors, draft contracts, compare pricing, and then stop dead because it has no way to pay for anything.

Mercury’s move makes the agent a recognized financial actor inside the company. Not an employee. Not a contractor. A new category with its own controls. The card has a limit. The limit is set by a human. Every transaction is logged and attributable. If the agent overspends or buys something unauthorized, it shows up the same way a rogue employee’s charges would — except the audit trail is cleaner because machines don’t lose receipts.

What matters here is the pattern. Agents are accumulating the same operational primitives humans have: identity, communication channels, spending authority. Each one independently seems minor. Together they describe something that looks a lot like a new kind of worker.