A transaction architecture for AI agents purchasing expert services

SSRN · Working paper ·

Develops an architecture and formal model linking delegated authority, payment and delivery of expert work. Analyses permission checks, repeated requests and transaction records, and maps the shift from managed expert calls to services that AI agents can purchase directly.

Read the paper on SSRN (46 pages)

The document

Title
An MPP-First, Settlement-Flexible Architecture for Self-Serve Agentic Expert-Access: Protocol Analysis, Formal Model, and Business-Model Transition
DOI
10.2139/ssrn.6928639
Abstract
SSRN 6928639
Posted
Length
46 pages
JEL classification
D47, L86, D82, E42

The problem it starts from

Autonomous and semi-autonomous software agents are becoming buyers of expert-derived work. The obstacle is not discovery or delivery but the transaction itself.

Established expert-access models route value through account management, manual scoping, scheduling and manual billing. None of those can be traversed by an agent, and each assumes a human at the point of purchase.

Its three contributions

  1. A layered reference architecture. Assigns each protocol in the agentic stack to its proper function, and states precisely what a payment gate does and does not replace.
  2. A formal model. A set-theoretic account of products, mandates, payment requirements and settlement, plus a finite-state machine for task fulfilment.
  3. A business-model transition. Maps packages, subscriptions and on-demand engagements onto self-serve, agent-addressable products, while preserving the manual exception path that high-stakes work requires.

Within the model, the paper proves four properties by construction: a gate-before-fulfilment safety invariant, idempotency under a request key, absence of double payment, and end-to-end auditability.

Where it is cited

The paper is the background citation in the Financial Stability Board, Reserve Bank of Australia and Bank of England responses. Each of those states that its arguments stand without it, and grounds its proposals in supervised payment practice instead.

Sources