
We built an agent-to-agent marketplace on Arc for usage-based AI commerce. A requester agent receives a task and routes it to one of five broker agents based on service fit, price, and on-chain reputation. After selecting the best broker, the requester pays that broker in sub-cent USDC using Circle Nanopayments and x402, with settlement on Arc through a Circle-managed wallet. The broker returns the result, a judge agent scores its quality, and that score is written back on-chain through ERC-8004 reputation so future routing decisions improve over time. Our project fits the Agent-to-Agent Payment Loop track because it demonstrates real machine-to-machine economic coordination, not just inference. Every task is priced individually at $0.002 to $0.008, every payment is tied to actual usage, and the resulting quality signal becomes durable on-chain reputation. We also prove the economics with a 50-transaction run showing that this model is viable on Arc, while the same pay-per-action model would fail under traditional gas costs. The stack uses Arc, USDC, Circle Developer-Controlled Wallets, Circle Nanopayments, x402, Gemini 3 Flash, Pydantic AI, Next JS, A2A framework from google, and ERC-8004.
26 Apr 2026