Artificial intelligence is dominating headlines with extravagant capital raising exercises and doom-laden proclamations from experts. Much about the technology’s future remains mysterious, but it is already clear that the payments world is likely to be dramatically reshaped by the arrival of agents.
At present, in many jurisdictions, agents are already assisting with commerce. Users ask for recommendations and AI sources products, compares offers and advises on the best purchase to make. Moving from here to a future where agents are able to complete purchases without immediate supervision from the authorising agents is a natural but challenging next step.
This edition of OMFIF’s Digital Monetary Institute Journal, together with a range of industry insights, examines what form this progress will take and what changes are necessary before it can happen.
Stablecoins may well represent an important vector of agentic payments, since the marginal cost of each payment is so low. David Anderson from Circle, one of the world’s most important stablecoin providers, warns that while many elements of the economy are ready now, ‘there are still critical pieces missing’.
Lara Calvo Lourido of Minsait warns that adoption of agentic payments depends on the implementation of strong controls, legal certainty and clear accountability. This question of accountability is key in agentic payments. Gillean Dooney from Barclays points out that agents’ arrival in payments is less about new payment rails and primarily about a gradual delegation of responsibility for financial decisions to AI.
While the arrival of agents powered by large language models is normal, Nilmini Rubin from Hedera points out that the last 60 years have been characterised by humans learning to trust machines to control money, beginning with automatic teller machines.
For this to take place in today’s economy, Giesecke+Devrient’s Philipp Edler argues a need for an ‘ecosystem of trust’ in which machines can identify themselves, transact and, crucially, be held accountable.
The arrival of agents in payments will mean that novel risks emerge also, and Biagio Bossone points out that the extreme pace at which agents operate and the instantaneous nature of the payments they authorise introduces an enormous challenge for the bodies that are supposed to supervise them.
Dom T. Ghazan from Global Trade Finance suggests that this problem will extend beyond regulators, affecting the legal and accounting systems of participants and payments services providers as well.
Pointing out the proliferation of agents in financial decision-making, Ben Miller and Thanos Alevizos from Agio Ratings argue this may also introduce the risks of flash crashes and runs as agents with the same basic model react instantly to the same information. Miller and Alevizos counsel that the economy will need to implement various risk layers to ensure that automation can be implemented safely.
Oracle’s Mark Rakhmilevich describes a future in which agents do much more than simply make retail purchases on behalf of individuals. He foresees a future where AI and blockchain combine to create a new operating model for the movement of money, with agentic payments taking place continuously, optimising liquidity distribution in real time, with other agents continuously monitoring for risks.
Chad Harper of Coinbase considers the policy implications of agentic payments, arguing, ‘Policy-makers should therefore focus less on fitting this market into existing categories and more on preserving the conditions it needs to operate.’
For Tram Anh Nguyen from the Centre for Finance, Technology and Entrepreneurship and Global Women in AI, we must examine how AI systems can inherit inequalities embedded in the data and records from which they emerge. If those patterns become entrenched, financial institutions could face an algorithmic monoculture in which AI increasingly shapes lending, risk assessment, supervision, capital allocation and fraud detection.
Sonja Davidovic and Hervé Tourpe of the International Monetary Fund also question whether policy will need to adapt, stating, ‘The decisive test for the financial sector will be whether an instrument can support automation while preserving legal certainty, oversight and reliable settlement.’
Interested in this topic? Subscribe to OMFIF’s newsletter for more.