Agent Pay

AgentPay enables AI agents to pay with MNEE using on-chain escrow, AI verification, and instant partial or full refunds—programmable money with built-in quality guarantees for autonomous systems.

Comment

Inspiration

As AI agents begin performing real economic work—writing code, generating content, analyzing markets, coordinating services—we realized there is no native way for them to transact safely and autonomously. Payments assume humans, disputes assume customer support, and refunds assume chargebacks. That entire model breaks for machines.

We were inspired by a simple question:
What does money look like when it’s built for autonomous agents instead of humans?

MNEE’s programmable, low-cost, and fast settlement properties made it the ideal foundation to explore this future.

What it does

AgentPay is an AI-native payment system that enables autonomous agents to:

  • Pay each other using MNEE
  • Escrow funds on-chain for tasks
  • Verify deliverables using AI
  • Automatically issue full, partial, or zero refunds based on quality

Funds are held in a smart contract. An AI verifier scores the delivered work, and the contract enforces settlement deterministically—no humans, no disputes, no chargebacks.

How we built it

We built AgentPay as a full end-to-end system:

  • Smart Contracts (Solidity)
    An AgentEscrow contract holds MNEE, manages task lifecycles, enforces deadlines, and executes settlement logic.

  • AI Verification Service
    A verifier agent evaluates task outputs against the original specification and produces a quality score from 0–100.

  • Agent SDK (TypeScript/Python)
    Enables agents to create tasks, approve MNEE transfers, submit results, and resolve escrow programmatically.

  • Demo + UI
    A terminal-based demo and frontend visualization show the full lifecycle: fund → deliver → score → settle/refund.

Crucially, the AI never controls funds—it only provides a score. All enforcement happens on-chain.

Challenges we ran into

  • Designing refund logic that works for machines, not humans
  • Ensuring the AI verifier was useful without introducing custodial or trust risks
  • Making the system simple enough to audit while still demonstrating real autonomy
  • Building a demo that clearly communicates a complex idea in minutes

Accomplishments that we're proud of

  • A fully autonomous agent-to-agent payment flo

Built With

Share this project:

Updates