Visa has created a programme to prepare the payments industry for AI agents that spend money on behalf of humans. Thredd, an AI-first payment processor, is the first European platform to join. Their customer Zilch will be among the first to let AI agents initiate payments.
This is not a niche experiment. Visa processes trillions of dollars in transactions annually. When Visa builds infrastructure for agent payments, it signals that AI-driven commerce is heading mainstream.
The way it works is straightforward. A consumer asks an AI agent to find a product within a budget. The agent searches, returns a recommendation, and the user approves the purchase with a biometric confirmation — a fingerprint or face scan. The agent then completes the transaction with the merchant.
From the consumer’s perspective, nothing fundamental has changed. They still authorise the payment. But the initiator is now software, not a person clicking “buy”.
Why existing payment infrastructure is not enough
Current payment systems assume a human is at the keyboard. Humans use passwords, two-factor authentication, and manual approval flows. AI agents cannot do any of these in the same way.
An agent cannot receive a text message with a one-time code. It cannot recognise a fraudulent transaction the way a person might. It needs different tools to establish trust.
Thredd’s solution involves three layers. First, tokenisation through Visa’s Token Service, so agents only ever handle payment tokens rather than real card numbers. Second, device binding that links tokens to specific trusted devices. Third, biometric passkeys that let a human authorise an agent-triggered payment with a fingerprint.
Building on this, Thredd is developing agent-specific tokens with custom permissions, plus fraud monitoring designed to catch patterns unique to agent behaviour — like execution drift, where an agent gradually deviates from its intended task.
What this means for merchants
If you run a business that accepts card payments, the shift to agent-initiated transactions matters. AI agents will behave differently from human shoppers. They will compare prices faster, process more options, and potentially place orders at machine speed.
Your payment systems need to handle this gracefully. That means supporting tokenised payments, handling high-velocity ordering without flagging it as fraud, and ensuring your checkout works without human interaction at every step.
The good news is that Visa and Thredd are building this infrastructure for the existing payments network. Merchants do not need to rebuild their systems. But they do need to be aware that an increasing share of transactions will originate from software, not people.
The bigger picture
Agentic commerce is still early. But the pieces are coming together. DoorDash built a command-line tool for agents. Visa is building payment infrastructure. Payment processors are developing agent-specific fraud rules. These are not isolated projects — they represent a coordinated shift toward a world where AI agents are economic participants.
For businesses, the message is consistent across all these developments: make your systems machine-readable, support automated transactions, and prepare for a customer base that includes both humans and software.
Practical takeaway
Visa’s Agentic Ready programme is a signal that agent-initiated payments are coming to the mainstream. The infrastructure is being built now. Businesses that prepare — by ensuring their payment systems work with tokens, support biometric authorisation, and handle automated ordering — will be ready when the wave arrives.
If you accept card payments, ask your payment provider about agent-ready capabilities. The question might feel premature today, but it will not feel that way for long. The infrastructure is being built, the standards are being set, and the first wave of agent-initiated transactions is already happening. Businesses that wait until the wave arrives will be catching up. Those that prepare now will have a head start on an entirely new category of commerce.