AI Agents Can Pay and Hire, But Not Trade Safely
Circle's MPP standardized payments, ERC-8183 standardized hiring, but AI agents still lack a trading standard; Hashlock Markets fills the gap with HTLC.
In June, Circle published the USDC-backed Machine Payments Protocol (MPP) specification for agentic payments, while a few months earlier Virtuals Protocol and the Ethereum Foundation's dAI team proposed ERC-8183 to standardize how agents hire one another, with BNB Chain already shipping a testnet implementation. In just a few months, two of machine commerce's three verbs - paying and hiring - have been standardized. The third verb, trading, still has no spec.
That gap isn't accidental. Payment is one-directional with bounded risk; hiring is subjective and needs judgment, which is why ERC-8183 built in an Evaluator role; but a trade is two-sided and objective - asset changes for asset, and the only risk is simultaneity. Forcing a trade through the payment spec recreates settlement risk, while forcing it through the hiring spec imports an unnecessary judge, bridge, or custodial trust assumption.
The requirements for a missing trade spec fall out naturally: price discovery without information leakage (sealed-bid RFQ), atomic execution where both legs lock to the same cryptographic condition (hash-time-locked contracts), automatic refund if a trade doesn't complete, asymmetric timeouts so legs don't expire simultaneously, and identity verification kept as a separate layer above settlement.
Examples like ClawBank's Manfred and Olas's Polystrat show agents are already trading, though mostly through custodial accounts or standing credentials. To close this gap, Hashlock Markets combines sealed-bid RFQ with HTLC settlement - live end-to-end on Ethereum mainnet, with Sui contracts in testing and Bitcoin validated on signet.