Bots that are supposed to buy

Payment systems spent years blocking fraudulent bots from impersonating customers. Now legitimate AI agents need to make real purchases, forcing a rethink of verification systems.

“We built really robust systems to verify that there's a human doing the transaction, 3DS, CVV codes, everything to prevent a bad bot from doing things.”

SolvaPay tackles agent payments

Stockholm-based fintech SolvaPay is building payment infrastructure specifically for AI agents, including machine-readable payments, usage-based billing, and spending controls.

The timing problem

AI agents developed faster than payment systems can adapt. Unlike the internet and smartphones, which took years to integrate into commerce, agentic AI is already operating autonomously.

“The timer already started a couple of years ago.”

One job, multiple payments

An AI agent can split work among other agents, creating chains of payments several layers deep. This complicates reconciliation when something fails partway through.

Agents as buyers and sellers

AI agents can purchase data and services, then resell what they build to other users. This requires usage-based billing for individual API calls and datasets rather than monthly subscriptions.

Connecting to existing rails

Agent payments must still flow through regulated financial systems with anti-money laundering checks, ledgers, and security controls. You can't create a separate payments environment for autonomous software.

“You can't skip steps.”

Merchants need new controls

Fraud systems must distinguish legitimate agent activity from attacks. Billing must handle per-unit purchases. Reconciliation must track multi-agent work chains. Payment choice could shift to the agent itself.