As AI agents start making payments for people and businesses, the questions of who the agent is and whether a transaction should go through are now answered in the same moment at authorisation. In this session, Ryan Dew, Chief Product Officer at Thredd, is joined by Anthony Craufurd from Visa, Steve Rayko from Zilch, and Stuart Barclay from Velocity to work through what that means for the network, the rail, the issuer, and the merchant.
Panellists: Ryan Dew (Thredd) | Anthony Craufurd (Visa) | Steve Rayko (Zilch) | Stuart Barclay (Velocity)
The panel covers where the hype is ahead of reality, why machine-to-machine payments are moving faster than consumer agentic shopping, how fraud prevention changes when settlement is final, what data an issuer needs before approving an agent-initiated transaction, and who carries the loss when an agent gets it wrong.
The new problem is delegated authority
Machines already initiate payments, so what is new is trusting software to make spending decisions on a person's behalf. That moves the issuer's question from who the customer is to whether the agent is acting within what the customer authorised.
Machine-to-machine payments are scaling first
Consumers still want a final approval step before an agent spends their money, while business payments and developers paying for compute and API calls already run on delegated authority and are moving faster.
Different transactions will run on different rails
A fraction-of-a-penny API call and a £900 flight need different economics and protections, so the panel saw cards and stablecoins working alongside each other, with stablecoins often improving settlement behind a card transaction.
Final settlement moves fraud prevention to the point of transaction
Card fraud controls rely on reversing payments after the fact, but on irreversible rails an agent could make hundreds of bad transactions before anyone notices, so trust has to be established before the payment goes through.
Knowing the agent is only half the decision
Issuers also need evidence of what the customer authorised, including limits on amount, merchant, geography, and timing, and that data has to arrive with the transaction within milliseconds. New agents with no history will need to borrow trust from an accountable party such as the network, a wallet, or the agent provider.
Autonomy should scale with risk
A £20 coffee order can run without interruption, but an unexpected £3,000 purchase, a new merchant category, or a new country should trigger a step-up or a decline.
'The agent got it wrong' covers several different failures
Wrong-item disputes currently follow existing e-commerce rules, but as agents gain autonomy, liability will depend on whether the fault sits with the customer, the agent, a failed control, or the merchant.
Ryan Dew, Chief Product Officer, Thredd
Anthony Craufurd, Director, Agentic Commerce & Growth Products, Visa
Steve Rayko, Senior Vice President of Engineering , Zilch
Stuart Barclay, Head of EMEA, Velocity
Speak to our team about what agent payments mean for your card programme.
