As companies hand more finance work to AI, the limiting factor may be permission rather than intelligence. That is the argument made by Ryan Taylor, senior vice president of product management for mobility and payments at WEX. He made the case in an interview for the PYMNTS B2B Payments Event 2026.

Intelligence is ahead of authority

Taylor says AI systems are getting better at spotting a suspicious invoice, flagging a supplier whose prices are creeping up, or deciding which transaction needs intervention. Knowing what should happen is different from having the authority to do it, especially when it means moving real corporate cash. In his view, that could turn payments from the last step in a business workflow into its control layer.

He says the problem today is that work happens in one system and payments in another, and people stitch the two together by hand. That causes delays and errors as volume grows. Payment data, he says, is the "digital footprint of business," showing not just what was spent but what was going on at that moment.

He gave a fuel example. A $100 charge is just an accounting entry. Add the vehicle's location, fuel level, route and past behavior, and an AI system can judge whether the purchase is legitimate and whether the driver could have fueled more cheaply elsewhere.

Virtual cards as "programmable permissions"

Taylor says effective AI needs context, controls and clarity, comparing it to onboarding a new employee. Virtual cards already let companies restrict merchant, amount, timing and number of transactions. In an agent-driven setup, those limits work as programmable permissions for machines.

"We don't want to give the agents open-ended access to our corporate bank accounts," he said. His model is that an agent decides an invoice should be paid, and the payment infrastructure decides whether it can pay it. That raises questions about which agent is acting, for whom, with what delegated authority, and when a human must approve.

Controls move earlier

Finance teams have traditionally caught problems after the fact, during reconciliation. Pairing AI with real-time payment data could move fraud, exception and variance checks closer to the moment money changes hands. Taylor calls this "shrinking the distance between this insight to action."

He expects the biggest gains to go to firms that treat payments as core infrastructure, not to those that add chatbots to legacy systems.