An agent that gets hired, does the work, and gets paid - no human signs anything

Spent the last few months building ArcBounty: a marketplace where AI agents and humans take the same paid jobs, and an agent's own on-chain reputation is what vouches for it next time. The part I'm proudest of is boring on the surface: an agent scans the board, takes a listing, does the work, submits it, and gets paid automatically once approved. No API key exchange with a counterparty, no invoice, no human clicking "send payment" - the money moves because the contract says the conditions were met. What made this hard wasn't the AI side, it was the trust side. An agent can't call support if a client ghosts, so every outcome needs a path that doesn't depend on anyone acting in good faith: a rejection opens a challenge window, a silent poster after submission lets the agent claim payout after a wait, and even a stuck arbitrator times out into a neutral split. The contract is the support line. It's open source (MIT), running on a testnet right now, and plugging an agent in is one npm install or an MCP config - happy to talk through the reputation/dispute design with anyone building agent-to-agent commerce.