Mercury Spend lets businesses issue Agent Cards, virtual cards created by a human but usable autonomously by an AI agent to complete purchases end-to-end without someone approving each transaction. The company controls the budget, eligible merchants and spending categories. Transactions outside those rules are declined automatically, and the agent cannot raise its own limit.
“People have already been manually issuing our virtual cards to agents,” Mercury Co-Founder and CEO Immad Akhund told Fast Company The new product, he said, extends that process with better controls, including the ability to audit agent spending. Mercury says more than 300,000 entrepreneurs use its platform, generating more than $650 million in annualized revenue.
Ramp Is Building the Same Idea Through a Different Approach
Mercury is not alone. Ramp is issuing scoped virtual cards through its Visa partnership that external AI agents can use within policy guardrails set by a company, PYMNTS reported. Ramp also launched AI Token Spend Management in July to track what companies spend on AI subscriptions and API usage. That’s a distinct problem from agent-issued purchasing: AI-vendor spending flows through subscriptions and usage fees, rather than through an agent’s own transactions, PYMNTS reported separately.
More than half of businesses, 52%, already use AI to manage part of their expense process, most commonly fraud detection and receipt capture, according to an American Express Trendex survey cited by PYMNTS. Another 40% plan to.
That backdrop makes a dedicated agent card a logical next step: Companies are already comfortable letting AI touch expenses. Giving an agent its own card extends that from monitoring spending after the fact to authorizing it in real time.
A Corporate Card Program Now Has Two Kinds of Cardholders
A human cardholder can be reasoned with or asked why a charge looks unusual. An agent cardholder either has authority to complete a purchase within its limits or it does not, enforced automatically at the point of sale. Mercury’s design reflects that: A human must create the Agent Card, and the agent can’t bypass or raise its own limit. That is narrower autonomy than a human employee has, since a person can ask for an exception while an agent’s card simply declines.
The practical test is whether routine, low-dollar agent purchases can move through existing card infrastructure without finance teams building new approval workflows for each one. Ramp’s own data offers an early answer: Across 50,000 businesses, companies using agents for expense management saw out-of-policy spend events fall 62% and policy flag rates drop 60% over two years, with its Policy Agent screening every transaction at 99% accuracy and declining out-of-policy purchases at the point of sale, PYMNTS reported.
If that pattern holds as agent cards spread beyond early adopters, spend management platforms will have added a second category of cardholder to a system built for one: AI agents will generate transaction records no different in form from a human employee’s, just faster and at a volume no manual review process was built to handle.
For all PYMNTS AI coverage, subscribe to the daily AI Newsletter.
