White paper
When agents become the customer
For 30 years, the payments industry has optimized for one action: persuading a human eye to tap a button on a glass screen. That audience is leaving the room.
Autonomous agents do not look at card art or care about brand loyalty. They evaluate net yield, FX markup, settlement speed and dispute terms in milliseconds. At the same time, the AI systems powering these agents require continuous, sub-cent settlement for model inference and compute, transactions existing card rails cannot economically touch.
This position paper outlines the structural shifts financial institutions must make to stay relevant, compliant and profitable as machine-led commerce arrives.
Five things that break – and what replaces them
When systems shift from human persuasion to machine arithmetic, traditional payment rails fail under the strain. Here is what breaks and what you need to build instead:
Brand-led selection.
Agents compare interchange, settlement and reward yield programmatically. If your terms are not machine-readable endpoints, your products will simply be omitted from the comparison.
Human-present authentication.
One-time passcodes and biometric prompts assume a person is at the keyboard. Agentic commerce requires durable, cryptographically signed mandates that prove delegated authority hours or days later.
Customer relationships.
Generalist AI platforms threaten to turn merchants into anonymous fulfillment hubs and isolate issuers from transaction context. You must turn payments into identity-carrying protocols to keep the customer relationship intact.
Core banking throughput.
Linear human shopping gives way to thousands of machine queries, micro-authorizations and rate checks in seconds. Batch systems and legacy fraud stacks buckle without burst-tolerant, idempotent architecture.
Fixed-fee economics.
Millions of sub-cent tool calls, token queries and compute reservations cannot bear a $0.50 card rail cost floor. Machine workloads require streaming, metered settlement.
What you will find inside the paper
Produced for Sibos 2026 in Miami under the theme Digital finance for AI-driven economies, this paper delivers an architectural and commercial roadmap for engineering teams and steering groups:
| Two rebuilds, one control plane: | Why front-end agentic commerce and back-end AI-stack settlement share the same architectural foundation. |
| The standards landscape: | Where emerging specifications – including MCP, ACP, AP2, Visa TAP and x402 – sit across the stack, and why agent identity remains unresolved. |
| Three execution imperatives: | The sequence to follow over the next 90 days, four quarters and 24 months to prove, ship and commercialize machine-speed trust. |
| The payments steering group checklist: | Eight diagnostic questions every executive team should ask to determine whether their roadmap is real or theoretical. |
If the buyer is an algorithm, what does it read about you?
We are hosting dedicated working sessions at Sibos 2026 in Miami Beach (September 28 – October 1).
Bring one of your core product lines. Together, we will trace its journey through the agentic transaction path — from initial discovery and machine evaluation to delegated mandate and final settlement. We will show you where your product risks being excluded, where authority checks break down and how to rebuild the architecture with defensible unit economics.
Meet the authors
