The Architecture Layer
Design patterns and reference architectures for systems where AI agents transact autonomously. Escrow, settlement, billing models, and revocability — the structural decisions that matter before you write a line of payment code.
The agent payment stack: a reference architecture
An opinionated end-to-end reference architecture for agentic payment systems — from the agent decision layer and tool calls, through authentication and authorisation, down to settlement and reconciliation. With a worked example using Bankee's own stack.
Escrow primitives for agent-to-agent transactions
When Agent A hires Agent B to complete a task, payment must follow verified completion — not just invocation. This post covers smart contract escrow patterns, optimistic release, dispute windows, and how to implement agent-to-agent escrow without requiring a blockchain for every transaction.
Micropayments vs session billing vs subscription — which model fits which agent use case
A decision framework for how to charge (and pay) in agentic systems. The right billing primitive depends on task duration, value predictability, and counterparty trust. We work through the trade-offs and map common agent patterns to the right model.
Designing for revocability: how do you stop an agent mid-spend?
Human-in-the-loop is a safety concept — but at the payment layer it has specific engineering requirements. This post covers kill switches, spend pause mechanisms, in-flight transaction cancellation, and how to design revocability without sacrificing throughput.
Reference architecture drops first
The full agent payment stack — decision layer to settlement — as an opinionated reference you can fork and adapt. Get notified when it's live.