Circle says agentic commerce still lacks trusted discovery

Circle's August 12 roadmap puts x402 and stablecoins at the payment layer and calls trust primitives less mature. Its test for an open agent market is one a closed platform cannot pass.

By XAgent Team · 2026-08-17

A payments company published a roadmap on August 12 that spends most of its length on what its own layer does not yet solve. Circle's Building the Open Agentic Economy places x402 and stablecoins at the value-transfer and settlement layers, then says the layer above them is not ready: "Trust primitives are less mature." It closes with an acceptance test that most agent platforms shipping today would fail.

That is a more useful document than another launch announcement, because the gap it names is the one merchants actually hit.

What the stack diagram concedes

Circle frames the problem as a missing protocol suite: "The agent economy needs its own stack of open internet protocols and blockchain-based trust primitives for identity, reputation, validation, and settlement." Their table maps each layer to a web equivalent — identity to "ERC-8004 (agentId + agentURI)" against "DNS + TLS certificates", value transfer to "x402 / MPP" against "HTTP: information transfer". The mapping is theirs, not our summary of it.

Two of those layers are shipping. On the payment layer they write: "x402 is designed to activate that missing payment layer, while stablecoins provide settlement." Anyone who has run an x402 flow knows the mechanics work; the protocol moved fast enough this month that client defaults changed underneath integrators.

The layer above is where the concession sits. On identity and reputation: "ERC-8004 defines registries for identity, reputation, and validation" — but "registries are only containers; credible evidence grounded in real behavior is still emerging." And on the layer that would let a buyer choose: "The capstone, trusted discovery, is still missing."

A registry that lists agents is not a market. It is a phone book with no way to tell which numbers answer.

The stranger test is the right bar for an open execution market

The sharpest line in the post is its acceptance criterion: "We will hold this milestone to the stranger test: the buyer must be a genuine third party in real traffic. An open market is proven when participants with no prior relationship can find and transact with one another."

We would put it more bluntly, because it is the distinction that decides whether any of this is a market at all. Almost every agent-commerce demo to date is a transaction between two parties who already knew each other — the same company on both ends, or two partners who integrated ahead of time. Those demos prove the plumbing. They do not prove the market, because the hard parts of a market are precisely the ones a prior relationship lets you skip: finding a counterparty, deciding whether to trust them, and pricing that risk.

Circle names the condition it wants to escape: "Today, many capable, specialized agents remain siloed inside companies or tied to closed platforms." A closed platform cannot pass the stranger test by construction. If the buyer and seller are both onboarded by the same operator, the operator is the trust layer, and the question of how strangers transact was never asked.

This is the thesis we have argued from the beginning of this blog, and it is why we describe what we are building as an open execution market rather than a platform. Open is not a licence choice. It is the condition under which the discovery problem is real rather than delegated.

Reputation has to come from transactions, not profiles

Circle's proposed mechanism is the part worth arguing with, because it is correct and harder than it sounds. They want "ranking grounded in observed behavior, not self-description", and describe the loop as: "Evidence from the completed transaction contributes to reputation that informs discovery for the next buyer."

That loop has an ordering problem. Reputation grounded in completed transactions requires completed transactions, and a buyer needs a reason to transact before any reputation exists. Every open marketplace has faced this; none solved it by publishing a registry schema. They solved it by making the first transactions cheap, bounded, and recoverable — small amounts, clear scope, a path to unwind — until enough behaviour accumulated to rank on.

Which puts a specific demand on the execution layer, and it is not a demand ERC-8004 or x402 can meet on their own. Identity says who an agent is. Payment says how value moves. Neither says what happens when a stranger's first order goes wrong — who holds the funds meanwhile, what the receipt proves, how a failed delivery is distinguished from a refused one. That is operations, and it is the layer where an open market either works or quietly becomes a closed one because only the well-connected can afford the risk.

What's next

The useful thing about the stranger test is that it is checkable. One question now separates plumbing from market when a vendor announces a milestone: was the buyer a genuine third party in real traffic, or the same company on both ends?

We would like more of the industry to publish against that bar, including ourselves. It is a harder number to report than transaction counts. It is also the only one that indicates whether agents are transacting with strangers yet. Building a service that agents can find and pay without a prior relationship is the problem we are working on. The discovery half of it is, as Circle says, still missing.