The calls

By 2030, the B2B AP function is one human and one agent. The agent runs settlement, the human handles exceptions.

What would prove it: by 2030, the standard mid-market AP team is one person and one agent, with the agent running settlement end to end.

Orchestration wins. Single-rail middleware does not. The losers are any platform that stops at workflow and never reaches the rail.

What would prove it: single-rail AP middleware loses share or exits while multi-rail orchestration platforms take the category.

AI agents can initiate the payment. The settlement layer still expects a person. That gap is about to become a payments infrastructure problem.

What would prove it: a card network or regulator ships agent-native authorization credentials after the liability gap surfaces in a real dispute.

Card networks opening their rails to AI agents marks the start of agentic settlement. Most payments people missed it.

What would prove it: agent-initiated transactions become a reported volume line at a major card network.

Reviewed 2026-08-01 · Visa and Mastercard both talked agents all earnings call and reported no agent volume line. Still open, watching every quarter.

Service-as-software replaces SaaS in the back office, payments first.

What would prove it: back-office buyers pay for completed outcomes instead of seats, and payments flips before the rest of the stack.

Reviewed 2026-08-01 · Customer service, not payments, is leading outcome pricing so far. The call is under pressure and still open.

AI won't earn software margins. It will earn payments margins.

What would prove it: AI-native back-office companies monetize on basis points of payment flow, not license fees.

Every AP vendor is betting on the wrong control point. Invoice processing is not the bottleneck.

What would prove it: invoice-automation pricing compresses while platforms that own payment execution hold theirs.

Access to AI has been democratized. The capacity to wield it has not. And the gap is widening faster than any technology gap in human history.

What would prove it: the productivity spread between heavy-AI operators and everyone else keeps widening in the data instead of converging.

The next internet monopoly will start at the payment layer.

What would prove it: a company that owns the payment layer for agent commerce compounds into monopoly-scale share.