The Autonomous Economy: How PayArk Becomes the Settlement Rail for Agent-to-Agent Commerce

The internet's commercial layer was built for humans. Every payment flow assumes a person is present: a browser renders a checkout page, a user enters card details, a merchant redirects to a return URL. Even the most sophisticated financial APIs, Stripe, PayPal, Razorpay, orient their UX around the assumption that a human is making a decision somewhere in the chain.

That assumption is collapsing.

AI agents are beginning to transact with each other autonomously. An AI agent managing your business's cloud infrastructure should be able to procure compute resources without human intervention. An AI agent managing your SaaS subscriptions should resolve past-due invoices automatically. An AI agent coordinating a supply chain should negotiate and settle contracts at machine speed.

The financial infrastructure to support this does not exist from Nepal's rails. PayArk is building it.

1. The Problem: Payment Infrastructure Designed for Browsers, Not Agents

Current payment flows have three properties that are fundamentally incompatible with autonomous agent commerce.

They require interactive user sessions. OAuth flows, 3D Secure challenges, and OTP confirmations are deliberate friction mechanisms designed to ensure human presence. They are good at this, and catastrophic for non-interactive autonomous systems.

They have no concept of delegated authority. When a human authorizes a payment, the proof of authorization is implicit, the person was there, they clicked the button, the browser session is evidence. When an agent initiates a payment on behalf of a human, there is no standardized way to prove the human actually delegated that authority, or that the agent is acting within its granted bounds.

They lack machine-readable trust signals. The entire ecosystem of fraud detection, behavioral biometrics, device fingerprinting, velocity checks, assumes a human pattern. Agents transacting legitimately look like noise because the signals don't map.

These are not edge cases. They are architectural failures that will become critical bottlenecks as agentic systems scale.

2. The Stack: A2A + AP2 + PayArk

Three emerging layers solve this problem in sequence.

Agent2Agent (A2A) is Google's open protocol for agent interoperability. It handles discovery: how does one agent find another, negotiate its capabilities, and establish a transport session? A2A uses Agent Cards, lightweight JSON documents published at .well-known endpoints, to advertise what an agent can do, what inputs it accepts, and what authentication schemes it requires.

The Agent Payments Protocol (AP2) sits above A2A and handles the cryptographic proof layer. It introduces Mandates, digitally signed verifiable credentials that create an immutable chain of authorization from a human principal through to a specific transaction.

PayArk is the settlement rail. Once A2A has established the agent channel and AP2 has produced the mandate chain, PayArk executes the actual payment against Nepal's financial carriers, eSewa, Khalti, ConnectIPS, with the same industrial-grade resilience and regulatory compliance we apply to standard merchant integrations.

Each layer does exactly one job. None overlap.

3. The Mandate Framework: Cryptographic Proof of Intent

The most consequential innovation in this stack is the Mandate. When a human buys a subscription, the proof of intent is behavioral. When an agent buys on behalf of a human, there is no behavioral trail. The agent needs a different form of proof, one that is cryptographic, precise, and non-repudiable.

Intent Mandate

Generated when a human authorizes an agent to act. It is a signed Verifiable Credential containing strict parameters:

Maximum transaction amount
Permitted vendor list
Execution time window
Agent identity

The human signs it with their private key. The mandate is proof that a real person, at a specific moment, granted a specific agent bounded authority to transact.

Cart Mandate

Generated at settlement time, inside PayArk's checkout flow, and only if the transaction's precise parameters fall within the Intent Mandate bounds. It links the signed human authorization directly to the specific transaction being settled.

This creates a closed chain: human intent → AP2 verification → PayArk settlement, with mathematically verifiable evidence connecting them.

If an agent attempts to exceed its mandate, charges above the ceiling, transacts outside the window, uses an unauthorized vendor, PayArk's mandate verification layer rejects the transaction before it reaches the carrier. The human's delegation is enforced at the infrastructure level, not the agent level.

4. What PayArk Already Provides

We've intentionally built PayArk's infrastructure to be ready for this moment.

A2A/AP2 Requirement	PayArk Status
MCP Server for agent interaction	✅ Shipped
`customer_id` as primary identifier	✅ Phase 4
Realtime Hub (Durable Objects)	✅ Shipped
HMAC-signed webhooks	✅ Shipped
SSRF egress protection	✅ Shipped
Idempotency keys	✅ Shipped
Effect-TS retry schedules	✅ Phase 6
NRB push-based model	✅ Architecture
Mandate verification layer	🔨 Phase 7
Agent Card endpoint	🔨 Phase 7

5. NRB Compliance Is a Feature, Not a Constraint

Nepal Rastra Bank's restrictions on automated pull deductions, often framed as a limitation, are architecturally superior for autonomous agent commerce.

Push-based subscription models require explicit customer initiation for each billing cycle. In an agent-first context, the Intent Mandate is the human initiation. Each billing cycle's mandate is a fresh cryptographic proof of authorized intent, satisfying both NRB requirements and the AP2 proof framework simultaneously.

PayArk's NRB-compliant push model isn't a workaround. It's the correct architecture for a regulated, non-repudiable autonomous payment system.

6. The Autonomous Economy

The endgame is a fully closed loop: AI agents that generate revenue through their services, receive payments into a PayArk-managed account, spend that revenue autonomously through mandate-governed transactions, and report clean financial records to their human principals, all without a human touching a payment flow.

An AI agent managing a SaaS product: charges subscribers via mandate-governed checkout flows, settles infrastructure costs autonomously, escalates to the human principal only when a transaction falls outside the mandate bounds.

An AI agent coordinating a supply chain: negotiates terms via A2A, generates Intent Mandates for approved vendors at approved price ceilings, executes settlement through PayArk when goods are confirmed delivered.

The financial infrastructure that makes this possible, mandate verification, non-repudiable audit trails, carrier-agnostic settlement, regulatory compliance, is exactly what PayArk provides.

We are not building a payment gateway for the agentic economy. We are building its central bank.

Links: