Agentic payments are payments made by an AI agent on your behalf. An AI agent is software that can carry out a task for you, not just answer a question. When you ask an agent to buy something, the agent also has to pay for it.
Imagine you tell an AI trip planner: book me a flight from Bengaluru to Delhi next Friday, under ₹6,000. The agent compares flights and picks one for ₹5,400. Then it reaches the payment page. But you are not there to enter an OTP or a UPI PIN. So the agent needs a way to pay, and the shop needs proof that you allowed the agent to pay.
Buying on someone's behalf like this is called agentic commerce. The payment part has three problems to solve.
First, the agent must not hold your real card number. Instead, it can be given its own token. A token is a different number that represents your card. The agent's token works only for the agent, and it has its own spending limit, such as ₹6,000. If the agent makes a mistake, only that token is cancelled.
Second, the shop must know that the agent is genuine. AI agent identity software gives each agent a verified identity, so that the payment gateway can check who is paying.
Third, there must be a record of what you allowed. For example: flights only, up to ₹6,000, until Friday. These limits are called AI agent guardrails. If there is a dispute later, the record shows whether the agent stayed within them.
This is still being built. Card networks, payment gateways and NPCI are testing ways for agents to pay by card and by UPI. No common method has been agreed yet.
Agentic payments will matter most for tasks that repeat. For instance, an AI shopping assistant could reorder groceries every week. A household assistant could pay the electricity bill. In each case, the agent pays, but you set the rules.