Everyone is building a "settlement layer." There are two of them.

#mcp #ai #cryptocurrency #blockchain

This is the Sunday recap — a step back from the week in agent-economy infrastructure, and where our own week fit into it.

The headline event: an $8M pre-seed round, led by a well-known fund, for a company describing itself as "the settlement layer for the agentic economy." That is a lot of money and a lot of conviction pointed at a thesis we have been building on for months — agents will transact autonomously, and they need infrastructure to do it. When a serious fund writes that check, it is a useful signal: the market is real, and we are not the only ones who think so.

But "settlement layer" is doing a lot of work in that sentence, and it is worth slowing down on. Because there are two different layers that both get called "settlement," and conflating them makes the whole agent-economy stack harder to reason about than it needs to be.

Layer one: the payment rail

The first layer answers a concrete question: how does an agent pay for something?

An agent books a flight, subscribes to an API, tips another agent for a result. On one side is the agent's wallet; on the other is a merchant who wants to be paid in a form they can use. The job of a payment rail is to wire those two together — handle the request, move the value, confirm the merchant received it. The x402 pattern (a payment embedded directly in an HTTP 402 response) is the cleanest expression of this idea, and it is growing quickly: the protocol is now clearing on the order of 500,000 transactions a week.

This is real infrastructure with real traction. The round this week is, in large part, a bet on this layer — agent-to-merchant payments, plugged into existing commerce. It is the work of plugging an agent into the economy that already exists.

It is just not the layer we build.

Layer two: trustless settlement

The second layer answers a different question: how do two agents exchange value when neither one trusts the other — and neither trusts an intermediary?

This is not "agent pays merchant." It is "agent A holds ETH, agent B holds BTC, they have agreed a trade, and now both legs have to happen — or neither." There is no merchant relationship here, no repeat business to lean on, no reputation either side can hold hostage. There is just a trade between two strangers across two chains.

The naive answer is to route through a custodian or a bridge: someone holds both sides, performs the swap, hands the results back. That works, and it is convenient, but it reintroduces exactly the thing an agent has no good way to evaluate — a trusted third party. A custodian can freeze funds. A bridge's lock-and-mint contract is a standing honeypot; the value sits there, waiting.

The trustless answer is a hash-time-lock contract (HTLC). Agent A locks funds against the hash of a secret. Agent B locks the corresponding funds on the other chain against the same hash. When A reveals the secret to claim B's funds, the act of revealing it lets B claim A's funds. Either both legs settle, or — after a timeout — both refund. No custodian ever holds both sides. There is no intermediate state where value sits exposed.

That is the layer we build. It is a primitive, not a product feature: a way for any two parties to perform a conditional value exchange across any two chains without trusting an intermediary.

They are a stack, not rivals

Here is the part worth being precise about. The payment rail and the settlement layer are not competitors fighting over the same job. They sit at different heights in the same stack.

An agent can use an x402-style rail to pay a merchant for a service — that is the rail doing its job well. The same agent can use a hash-time-lock to settle a cross-chain trade with another agent — that is the settlement primitive doing a different job. One is "buy something from the existing economy." The other is "exchange value with a peer, trustlessly." A mature agent will do both, often inside the same workflow.

So when this week's round gets described as funding "the settlement layer for the agentic economy," the honest reading is: it funded a payment rail, and payment rails are valuable. The trustless cross-chain atomic settlement layer underneath — no custodian, no bridge — is a different layer, and it still needs building. That is the part we have spent the week on.

What our week looked like

Pull our own threads from the past seven days and every one of them turns out to be the same primitive seen from a different angle:

Forward-dated settlement. An HTLC timeout is usually treated as a safety valve. Used deliberately, it is a delivery window: set it to 24 hours and an atomic swap becomes a T+24h forward — a price locked today, delivery enforced on-chain tomorrow, with no clearinghouse holding the asset in the gap.
Sealed-bid quote requests. A quote request leaks direction, size and urgency the moment it is broadcast. A sealed-bid RFQ removes the pre-trade leak — makers bid blind — and atomic settlement removes the post-commit leak, because there is no gap between "agreed" and "settled" for a last look to live in.
Counterparty filtering. Trustless settlement removes the need to trust a counterparty's intent. Agents still benefit from filtering on a counterparty's history — on-chain data, not identity claims.

Three different-looking features. One hash-time-lock underneath all of them.

The usual honest caveat

Because we keep this straight: atomic settlement is live end-to-end on Ethereum mainnet today. The Sui contracts are deployed and CLI-tested, with gateway wiring still in progress. The Bitcoin P2WSH HTLC is validated on signet, with mainnet pending. When this recap says "live," it means Ethereum mainnet — the rest is honest roadmap.

And HTLCs are not free. They cost capital lockup for the duration of the timeout, they carry timeout-coordination risk across chains with different block times, and they assume the hash function holds. We think those are the right tradeoffs for trustless settlement. They are still tradeoffs, and we would rather say so than pretend otherwise.

The question for the week

So here is what we would leave you with. The agent economy clearly needs payment rails — the funding this week says so, and the transaction counts back it up. But it also needs a layer below the rails: a way for agents to settle with each other directly, with no one in the middle.

When you read "settlement layer for the agent economy," which layer do you actually picture — the rail that plugs an agent into merchants, or the primitive that lets two agents trade without trusting anyone?

Hashlock Markets — atomic settlement for the agent economy. Sealed-bid RFQ + HTLC settlement, fused into one operation. No bridges, no custodians.